This achievement and other advances, which Qubit says raise hopes of a near-term practical application of hybrid high performance computing (HPC)-quantum computing to drug discovery, has led to the company and the Sorbonne receiving €8 ($8.7) million in funding under the France 2030 national plan for the further development of Hyperion-1.

With this proof of concept, the teams note that they have demonstrated that the routine use of quantum computers coupled with high-performance computing platforms for chemistry and drug discovery is much closer than previously thought. Nearly five years could be gained, they add, bringing researchers significantly closer to the era when quantum computers (noisy or perfect) could be used in production within hybrid supercomputers combining HPC, AI, and quantum. The use of these new computing powers will improve the precision, speed, and carbon footprint of calculations, the researchers point out.

Soon to be deployed on noisy machines

To achieve this breakthrough, teams from Qubit Pharmaceuticals and Sorbonne University developed new algorithms that break down a quantum calculation into its various components, some of which can be calculated precisely on conventional hardware. This strategy enables calculations to be distributed using the best hardware (quantum or classical), while automatically improving the complexity of the algorithms needed to calculate the molecules’ properties. In this way, explain the researchers, all calculations not enhanced by quantum computers are performed on classical GPUs.

As the physics used allows the number of qubits required for the calculations, the team, by optimizing the approach to the extreme, has managed to limit GPU requirements to a single card in some cases, according to the scientists. As this hybrid classical/quantum approach is generalist, it can be applied to any type of quantum chemistry calculation, and is not restricted to molecules of pharmaceutical interest, but also to catalysts (chemistry, energy) or materials, notes Robert Marino, PhD, CEO of Qubit Pharmaceuticals.

“This work clearly demonstrates the need to progress simultaneously on hardware and software development,” says Jean-Philip Piquemal, PhD, professor at Sorbonne University and director of the theoretical chemistry laboratory (Sorbonne University/CNRS), co-founder and CSO of Qubit Pharmaceuticals. “It is by making breakthroughs on both fronts that we will be able to enter the era of quantum utility for drug discovery in the very short term.”

The post Qubit Number to Simulate Molecules Reportedly Reduced by the Sorbonne and Qubit Pharmaceuticals appeared first on GEN - Genetic Engineering and Biotechnology News.

The company is on track for an early 2025 launch of a first-in-human Phase I trial of lead candidate ABS-101, CEO Sean McClain told GEN Edge, with an interim data readout expected in the second half of next year.

Absci is in IND-enabling studies for ABS-101, which is designed to treat inflammatory bowel disease (IBD) by targeting tumor necrosis factor-like cytokine 1A (TL1A). The company began those studies earlier this year after selecting a primary and a backup development candidate from three versions of ABS-101.

The primary candidate was selected based on its profile seen in mouse PK studies, the potency it showed in in vitro assays, and its binding affinity, as well as its manufacturability and developability.

“The candidate that had the best attributes across the board is the one we advanced as the actual candidate,” McClain said. “The backup had the same potency, similar affinity, and similar half-life. There were just some attributes on the developability side that weren’t as good as the actual candidate itself. But it still was a strong candidate.”

In releasing first-quarter results last month, Absci also presented preclinical data showing that all three versions of ABS-101 could bind both the TL1A monomer and trimer—activity that the company said could potentially lead to differentiated clinical efficacy.

“This is really exciting because it allows us to potentially—and we’ll find it out in the clinic—be able to go after patient populations that competitor molecules are unable to,” McClain said.

All three versions of ABS-101 demonstrated high affinity, high potency, favorable developability, and extended half-life. In addition, using its de novo AI model, Absci has designed ABS-101 toward a specific epitope with the aim of achieving superior potency and lower immunogenicity.

At the 42^nd Annual J.P. Morgan Healthcare Conference in January, Absci presented preclinical data on ABS-101 from multiple biophysical and cellular assays showing it to demonstrate equal or superior potency to two other TL1A-targeting drug candidates—RVT-3101, Roche’s TL1A-directed antibody candidate being developed for IBD, including ulcerative colitis (UC) and Crohn’s disease (CD); and Tulisokibart (MK-7240), which Merck & Co. is developing to treat immune-mediated diseases, including UC, CD, and other autoimmune conditions.

Later this year, Absci plans to share additional preclinical data, including results from nonhuman primate studies of ABS-101, likely via press release and data package.

Two additional candidates

Also in Absci’s pipeline are two additional preclinical candidates. One is ABS-201, a lead optimization phase drug to be developed for an undisclosed dermatological indication that according to the company has significant unmet need.

“I will say that we view this as an underappreciated target similar to TL1A,” McClain said.

TL1A was underappreciated, he said, until 2022 when Tulisokibart, then known as PRA023, generated positive clinical data as the lead candidate of Prometheus Biosciences—which Merck acquired last year for $10.8 billion.

“We think this derm target [ABS-] 201 is a very similar type of profile,” McClain asserted.

Absci expects to have a primary candidate for ABS-201 by the end of the year, at which point the company plans to release preclinical data. The company says ABS-201 has potential to be a best-in-class antibody showing high efficacy following once monthly or less frequent, low-volume, subcutaneous injection.

The other preclinical candidate in Absci’s pipeline is ABS-301, a lead optimization phase, fully human antibody that is designed to bind to a novel target discovered through Absci’s Reverse Immunology platform.

“We took a patient sample that had an extraordinary immune response, took the antibodies from the tertiary lymphoid structure, and then did a protein panel screen to find out what antigens or targets it was binding to. And that ended up allowing us to discover ABS-301, a novel I/O target,” McClain said.

He said the mode of action for ABS-301 will be validated in in vivo studies that are expected to be completed in the second half of this year: “This will be like essentially preclinical proof-of-concept similar to a Phase II POC [proof-of-concept].

Absci considers ABS-301 a potentially first-in-class immuno-oncology candidate. ABS-301’s target, which has not been disclosed, is an immune evasion strategy to limit immune infiltration and turn tumors immunologically “cold.” Absci reasons that ABS-301 treatment in cancer may release immune suppression and permit immune cells to infiltrate the tumor, allowing for a robust anti-tumor response.

Oncology and immunology and inflammation

Oncology and immunology and inflammation (I&I) have emerged as Absci’s key therapeutic areas of interest. In addition to -101, -201, and -301, Absci said, it expects to advance at least one additional internal asset program to a lead-identification stage this year.

The company’s pipeline could also grow if it succeeds in growing current collaborations with biopharma partners and establishing new ones. Absci has said it expects to sign drug creation partnerships with at least four additional partners this year, of which at least one could be a multi-program partnership.

Absci continues to make further progress on its existing drug creation partnerships and anticipates signing additional drug creation partnerships with at least four partners in 2024, including one or more multi-program partnerships.

“We are on track to achieve that goal,” McClain said. “We look to partner anywhere from a drug candidate phase, where we have preclinical proof of concept, all the way to a Phase II. There are a lot of factors that end up going into that, like what are the costs of the clinical trials? What is our own internal domain expertise? I think something like ABS-301, where we don’t have as strong of I/O expertise, that may get partnered sooner rather than later, but there’s a lot of factors that go into when we go about partnering.”

Absci finished the first quarter with $58.831 million in cash, cash equivalents, and another $102.712 million in short-term investments—enough of a “runway,” the company says, to fund its operations into the first half of 2027.

For the full fiscal year, Absci said, it expects to use approximately $80 million in cash, cash equivalents, and short-term investments—including expected costs associated with completing the IND-enabling studies for ABS-101 with an undisclosed contract research organization.

The post Absci Eyes Busy 2025 as Lead AI-Designed Candidate Heads for Clinic appeared first on GEN - Genetic Engineering and Biotechnology News.

The study is published in PNAS, in the paper, “Pairing interacting protein sequences using masked language modeling.”

DiffPALM leverages the power of protein language models, an advanced machine learning concept borrowed from natural language processing, to analyze and predict protein interactions among the members of two protein families with unprecedented accuracy. It uses these machine learning techniques to predict interacting protein pairs. This leads to a significant improvement over other methods that often require large, diverse datasets, and struggle with the complexity of eukaryotic protein complexes.

Another advantage of DiffPALM is its versatility, as it can work even with smaller sequence datasets and thus address rare proteins that have few homologs. It relies on protein language models trained on multiple sequence alignments (MSAs), such as the MSA Transformer and AlphaFold’s EvoFormer module, which allows it to predict the complex interactions between proteins with a high degree of accuracy. Even more, using DiffPALM shows high promise when it comes to predicting the structure of protein complexes, which are intricate structures formed by the binding of multiple proteins, and are essential for many of the cell’s processes.

In the study, the team compared DiffPALM with traditional coevolution-based pairing methods, which study how protein sequences evolve together over time when they interact closely. This is an extremely important aspect of molecular and cell biology, which is well-captured by protein language models trained on MSAs. DiffPALM is shown to outperform traditional methods.

The application of DiffPALM is obvious in the field of basic protein biology, but extends beyond it, as it has the potential to become a powerful tool in medical research and drug development. For instance, accurately predicting protein interactions can help understand disease mechanisms and develop targeted therapies.

The researchers have made DiffPALM freely available, hoping that the scientific community adopts it widely to further advancements in computational biology and enable researchers to explore the complexities of protein interactions.

The post AI-Based Approach, DiffPALM, Helps Determine Protein-Protein Interactions appeared first on GEN - Genetic Engineering and Biotechnology News.

In their paper, which is titled, “Interpreting cis-regulatory mechanisms from genomic deep neural networks using surrogate models,” the developers explain that SQUID uses “simple models with interpretable parameters” to “approximate the DNN function with localized regions of sequence space.” They claim that unlike other methods, SQUID “removes the confounding effects that nonlinearities and heteroscedastic noise in functional genomics data can have on model interpretation.” As proof of its effectiveness, they present results from experiments that show that SQUID “consistently quantifies the binding motifs of transcription factors, reduces noise in attribution maps, and improves variant-effect predictions.”

“The tools that people use to try to understand these models have been largely coming from other fields like computer vision or natural language processing. While they can be useful, they’re not optimal for genomics,” explained Peter Koo, PhD, an assistant professor at CSHL and senior author on the paper. “What we did with SQUID was leverage decades of quantitative genetics knowledge to help us understand what these deep neural networks are learning.”

SQUID works by generating an in silico library of variant DNA sequences, training a surrogate model called a latent phenotype model on the data using a program called Multiplex Assays of Variant Effects Neural Network or MAVE-NN, and then visualizing and interpreting the model’s parameters. With this tool, scientists can run thousands of virtual experiments simultaneously and identify which algorithms make the most accurate predictions about the variants. 

While virtual experiments can’t exactly replace laboratory tests, “they can be very informative” for helping scientists form hypotheses for how a particular region of the genome works or how a mutation might have a clinically relevant effect,” said Justin Kinney, PhD, a CSHL associate professor and one of the co-authors of the study. 

The scientists also described using SQUID to study epistatic interactions in cis regulatory elements as a way to evaluate its performance. To test whether SQUID could work on this task, they “implemented a surrogate model that describes all possible pairwise interactions between nucleotides within a sequence.” They then used the model “to quantify the effects of pairs of putative AP-1 binding sites.” Their results demonstrated that the “pairwise-interaction models” they created yielded more accurate results than “additive surrogate models.” Specifically, SQUID was able to “quantify epistatic interactions that were otherwise obscured by global nonlinearities in the DNN.”

Compared to several other methods, SQUID is more computationally demanding, its developers noted. They suggest that it may work better for researchers working on in-depth analysis of specific sequences such as disease-associated loci rather than those working on large-scale genome analyses. 

The post New Computational Tool Elucidates How Deep Neural Networks Interpret Genomic Data appeared first on GEN - Genetic Engineering and Biotechnology News.

Insilico has moved the first dozen of an eventual 25 staffers to a 3,000-square-foot suite at 1000 Massachusetts Avenue. The company will retain operations in New York at the Cure, the Deerfield-owned life sciences and healthcare campus at 345 Park Avenue South.

“We have relocated ultra-high level pros, basically the brains of the operation,” Insilico founder and co-CEO Alex Zhavoronkov, PhD, told GEN Edge. “Most of the target selection, product selection strategy, business development strategy, toxicology, some high level strategic medicinal chemistry are going to be there—and of course, partnering.”

In an interview on the sidelines of the recent Biotechnology Innovation Organization (BIO) International Convention held in San Diego, Zhavoronkov said Insilico wanted to address one of its priorities, namely scaling up operations in the United States, within a region that has attracted major hubs for global biopharmas. Boston/Cambridge has been a longtime leader of GEN’s nationally-quoted A-List of Top 10 U.S. Biopharma Clusters, though last summer the region narrowly lost the top spot to the San Francisco Bay Area after a falloff in venture capital financings reflecting 2022’s post-pandemic bear market.

1000 Massachusetts Avenue is a four-story, 105,217-square-foot office building owned by Intercontinental Real Estate, a Boston-based real estate investment management firm.

Zhavoronkov said Insilico’s new headquarters will also include “a really nice showroom.”

“We are going to launch a few more software products and maybe, just maybe, hardware products which will allow you to compute on site. That’s why having a showroom is very important,” Zhavoronkov added.

The company has yet to disclose details on its planned future software and possible hardware products, though Zhavoronkov said the software will be a “confidential compute” offering designed to let biopharma users design drugs using AI but without storing their data in the cloud.

Software solutions accounted for just 6.6% or $3.362 million of Insilico’s total $51.18 million in revenues last year. Software revenues more than doubled year-over-year, zooming 124% from $1.499 million in 2022.

“My target for this year is substantial: At least 15% of sales has to be coming from software,” Zhavoronkov said. “I want to continue with software plus drugs, because we are not a biotech. Once you are going to just focus on biotech, you are no longer an AI.”

He said Insilico is looking to reposition its business development efforts by innovating on “really ultra-high quality data packages”: “Every program, particularly for the United States, needs to be of ultra superior quality so that when pharma looks at it, they need to be extremely impressed,” Zhavoronkov said. “When you are selling cookies to the cookie makers, they know how to make cookies, right?”

Fabergé eggs of AI

He likens his company’s development of AI drug candidates to the jeweled Fabergé eggs created by Peter Carl Fabergé’s House of Fabergé in Saint Petersburg, Russia. Between 1885 and 1917, nearly 70 such eggs were crafted, of which 57 survive—including 46 of the 52 “Imperial” eggs made as Easter gifts for the wives and mothers of Russian czars Alexander III and Nicholas II.

“We want to become a Faberge factory where we take the egg, design and properly decorate it, then go and take it for the show so people can bid for it,” Zhavoronkov explains.

Just as Faberge egg customers customize their selections with a jewel maybe some music, Insilico envisions customizing its generative AI drug candidates, then putting them up for bids to eager biopharma customers.

“We want to be the super high-end jeweler who puts all of those assets on display for purchase. Our job is, at the very least, to offer best in class chemistry,” Zhavoronkov said. “If anyone who is looking at the asset tells me, I actually saw a better molecule somewhere else in this target class, I would be very disappointed, and I might deprioritize the program.”

“I need to make sure that they are at least best in class and then for many, I have first in class, so there is nothing to compare. You’d have to compete on biology,” Zhavoronkov added. “We are taking those Faberge eggs to the prom, so to speak, right?”

Insilico has built a pipeline of 31 programs aimed at 29 targets. Seven programs have received IND approval from the FDA and advanced into clinical phases, starting with lead candidate INS018_055, a small-molecule TRAF2- and NCK-interacting kinase (TNIK) inhibitor that is in Phase IIa trials for idiopathic pulmonary fibrosis (IPF) in the United States (NCT05975983) and China (NCT05938920). The company expects to release its first Phase II data for INS018_055 later this year.

On Tuesday, Insilico announced that it completed enrollment of 71 patients at 29 sites across China in the Chinese Phase IIa study, designed to evaluate the safety and tolerability of oral INS018_055 for up to 12 weeks in adults with IPF compared to placebo. and that the company is preparing to launch a Phase IIb proof-of-concept study in 2025 to explore the efficacy and further safety of INS018_055.

The American Phase IIa study, designed to enroll 60 patients, is “ongoing and actively accruing patients,” Insilico’s chief medical officer Sujata Rao, MD, stated.

An inhalable form of INS018_055 for IPF is in the IND-enabling phase, as is a program to develop the candidate as a treatment for kidney fibrosis.

In a paper published in Nature Biotechnology, a team of 30 researchers led by Zhavoronkov detailed how they used generative AI to discover INS018_055, with a novel target discovered by Insilico’s target identification engine, PandaOmics, and a novel molecular structure designed by its generative chemistry engine, Chemistry42. Both are specific-function platforms within the company’s AI platform, Pharma.AI.

“Golden target”

“We’re constantly expanding the indications. I think that this is a golden target, and people should be looking at this not as fibrotic target, but also as a cancer target as well. We want to ensure that it gets established as a fibrotic agent or a cancer agent,” Zhavoronkov said.

Zhavoronkov led a team of seven researchers from Insilico and academic partner ETH Zurich in publishing a paper in Trends in Pharmacological Sciences concluding that TNIK was a suitable target for treatments against several aging-related diseases, including fibrosis, cancer, obesity, and Alzheimer’s disease.

“TNIK signaling appears to converge on four critical hallmarks of aging, cellular senescence, deregulated nutrient sensing, chronic inflammation, and altered intercellular communication. TNIK’s contribution to these processes implicate it as a possible contributor to aging-related pathology, particularly by promoting conditions like cancer and metabolic dysregulation,” the researchers observed. “An interesting hypothesis would be whether TNIK dysregulation contributes to the aging process itself or is a consequence of it.

How did Insilico come to identify several potential indications for TNIK?

“We used aging research as a platform for target discovery,” Zhavoronkov explained. “We didn’t just use AI. We were looking for common mechanisms that work in aging and disease. We think that if something works in multiple areas of aging, multiple hallmarks of aging, it should be powerful enough to work in multiple diseases. Now, independent groups have implicated TNIK in Alzheimer’s and neurodegeneration.”

Alzheimer’s has been a notoriously challenging disease for which to develop drugs; a landmark Cleveland Clinic study pegged at 99% the failure rate of drug candidates targeting the memory-robbing ailment. But Eisai and Biogen won FDA approval last year for the anti-amyloid antibody Leqembi® (lecanemab-irmb), and last week an FDA advisory panel recommended agency approval of Eli Lilly’s amyloid plaque-targeting therapy donanemab.

In obesity, INS018_055 would compete with a growing number of newer treatments led by glucagon-like peptide 1 (GLP-1) receptor agonists such as Novo Nordisk’s semaglutide, marketed for type 2 diabetes treatment as Ozempic® and for obesity as Wegovy®—as well as Eli Lilly’s tirzepatide, marketed as Mounjaro® for type 2 diabetes and as Zepbound® for obesity.

Last year a research team from University of Copenhagen and research partner institutions reported associations of TNIK variants with blood glucose, HbA1c, body mass index, body fat percentage, and feeding behavior in a paper published in Science Advances: “These results define an untapped paradigm of TNIK-controlled glucose and lipid metabolism.”

Using genetically modified Drosophila and mouse models, the researchers studied the role of TNIK in obesity and metabolic dysfunction. In experiments with mice, researchers eliminated the gene for TNIK, which enabled the mice not to gain weight, no matter how much high-fat food they were fed.

“Like Schwarzenegger”

“The beauty of this paper was that if you look at mice on high fat, high sugar diets, TNIK wild type, they look so muscly, like Schwarzenegger. Inhibiting TNIK makes them thin. Now we are figuring the biology, the why. We don’t know, nobody knows yet,” Zhavoronkov said “We are very, very happy that it worked on obesity in that group.”

Insilico’s most recent clinical candidate is ISM3412, a treatment for methylthioadenosine phosphorylase (MTAP) deleted cancers, which account for 15% of all cancers. In April, Insilico won FDA IND approval to begin a Phase I multicenter, open label study (NCT06414460) of ISM3412, a small molecule selective methionine adenosyltransferase 2A (MAT2A) inhibitor designed to target cancers with MTAP deletion. The study, which has not yet begun recruiting patients, is designed to assess the drug in patients with locally advanced and metastatic solid tumors.

Insilico is one of several companies focused on treating cancer by inhibiting MAT2A. Furthest along is Ideaya Biosciences with its Phase II candidate IDE397, while Servier has listed a MAT2A inhibitor, S95035, as being in Phase I/II development; a Phase I trial (NCT06188702) is now recruiting patients, according to ClinicalTrials.gov.

In addition to Cambridge, MA, and New York, Insilico has operations in:

• Abu Dhabi, UAE, one of the company’s two hubs for its Pharma.AI scientific team

• Hong Kong, which oversees Insilico’s R&D collaboration and software solution businesses.
• Montreal, the company’s other hub for its Pharma.AI scientific team, and a business development location.

• Shanghai, one of two sites where the company focuses on drug discovery, R&D collaboration, and business development.
• Suzhou, China, home to Insilico’s wholly owned, generative AI-driven automated laboratory, launched in December 2022.
• Taipei, the company’s other drug discovery and business development site.

Since its founding in 2014, privately held Insilico has raised more than $400 million, the company stated in an updated prospectus filed in March with the Hong Kong Exchange. The company has applied to the Exchange for approval to carry out an initial public offering (IPO). The company applied for a listing in June 2023 but allowed its application to lapse.

According to its updated filing, Insilico’s $51.18 million in 2023 revenue marked a 70% increase from the $30.147 million the company generated in 2022. Most of Insilico’s revenue, $39.022 million or 76%, came from pipeline drug development, which includes R&D and out-licensing of some drug candidates to biopharmas, while another 17% or $8.976 million came from drug discovery services, such as collaborations with biopharma partners to discover targets associated with diseases using Pharma.AI, identify those targets, and conduct R&D.

Insilico is one of two AI-based drug developers that have sought to go public through the Hong Kong Stock Exchange in recent months.

XtalPi launched its IPO on Friday, raising HKD $989.3 million ($126.8 million)—the third largest IPO raised to date this year on the Hong Kong exchange, and the first to list there under a new rule designed to attract “strategic” startups in industries that include biotech. Established in 2015 by three postdoctoral physicists at MIT, XtalPi also has a U.S. site in Cambridge, MA, as well as Chinese sites in Beijing, Shenzhen, and two in Shanghai.

“Leveraging our Pharma.AI platform and automated laboratory, we plan to advance our pipeline in a rapid and focused manner. We plan to utilize our network of over 40 CROs [contract research organizations] and CDMOs [contract development and manufacturing organizations] for our nominated preclinical candidates and advance those programs into the clinical stage,” Insilico stated in its IPO. “We expect to advance our internal discovery-stage programs and nominate four to five additional preclinical candidates in the next 12 months.

The post Insilico Moves HQ to Cambridge, MA, Completes IPF Trial Enrollment appeared first on GEN - Genetic Engineering and Biotechnology News.

Key challenges include process uncertainty and batch-to-batch variability, manufacturing dynamics, and addressing market pressure to meet growing demands at lower costs.

Transitioning “from science labs to smart operations…requires a proactive use of data analytics and operations management to inform daily decisions,” according to a recent paper by scientists from Eindhoven University of Technology (TU/e), MSD, Merck USA, GEA Group, and Ceva Santé Animale. The roadmap they developed begins with strategic vision and the right technology infrastructure, such as continuous manufacturing and real-time release testing.

The process intensification inherent in continuous manufacturing makes it attractive to biomanufacturers. Now, more sophisticated sensors are providing processing data that otherwise is difficult to obtain. Consequently, “Continuous manufacturing systems will generate large amounts of process data,” Tugce Martagan, PhD, associate professor, TU/e, told GEN.

Greater gains

To achieve greater gains, however, “We also need to equip these new technologies with operations management/AI-driven control algorithms to achieve optimal performance,” Martagan says. By combining in-depth sensor data with increasingly sophisticated analytics and operations management strategies, deeper, often cross-functional, insights into production processes may emerge that can further improve operations and results.

“It is important to have the right vision and technology infrastructure to collect, store, and make the most of the process data to achieve the best performance,” Martagan emphasized.

The most important optimization step, however, may be to standardize processes. “Our industry depends on so many known and unknown parameters, and, if known, their contribution to the whole is often not known well,” Bram van Ravenstein, associate director and operations lead for bacteriological production, MSD, told GEN. “By highly standardizing the process, gathered data can more easily be used to start building models.”

Furthermore, he noted, multidisciplinary expertise is vital not only to develop and implement the initial models, but to transfer knowledge among manufacturing sites.

“The easiest place to start are packaging and filling processes, because the number of contributing parameters is reduced,” van Ravenstein added.

The team’s original study involved tablet production, but, “I believe many of the insights, models, and conclusions can be transposed to other pharmaceutical processes, [including biotechnology], therefore boosting gains in other areas,” van Ravenstein said.

Ultimately, Martagan concluded, “We hope the roadmap and data we have presented will inspire and facilitate new research at the intersection of operations management, data science, and biomanufacturing to advance science and industry practice.”

The post Transitioning to Smart Operations appeared first on GEN - Genetic Engineering and Biotechnology News.

To achieve these dual goals, biopharmaceutical companies must be willing to use innovative bioprocessing technologies and embrace the human-centric ideas of the emerging “Industry 5.0” model, according to new research.

The study, by a team at the Sargent Centre for Process Systems Engineering at Imperial College London in the U.K., examined industry efforts to move toward net zero, socially sustainable, and eco-efficient operations.

Novel digital tool development

One key finding is that, although it is hard to quantify process sustainability at present, new digital tools are being developed to help drug companies better understand their environmental impact.

“Currently, it is challenging to pinpoint more or less sustainable products if we consider that processes are product-specific and also differ on a company-basis,” according to lead author Miriam Sarkis, a doctoral student. “The good news is that model-based tools to answer this question exist and have been developed to tackle case studies for other industries.”

“With the recent growing interest in sustainability within biopharma, we are surely going to see a larger volume of comparative studies screen production pathways for a range of products.”

Quantification difficulties aside, the researchers also identified some other common challenges. In upstream processes, for example, the major issue is the large amount of water needed for cell cultivation. For downstream processes, the requirement for resins and solvents is the biggest difficulty from a sustainability standpoint.

“In addition to this,” said Sarkis, “if manufacturing relies on stainless steel equipment additional water is required to sterilize and clean equipment after each batch. Switching to single-use equipment has been shown to help reduce the water requirements for inter-batch cleaning. The industry has seen an increasing uptake of single-use technologies driven by a need to build more flexible and easily scalable facilities.”

Continuous manufacturing

Another finding was that growing industry interest in new production models is, in part, driven by a desire to make processes more sustainable.

“Process intensification through continuous manufacturing helps lower costs and environmental footprint per unit product and is surely the sustainable way forward for the sector,” Cleo Kontoravdi, PhD, professor and co-author of the paper, told GEN.

But companies looking to make processes more sustainable through intensification still face some technical and regulatory challenges.

“At regulatory level, there seems to be a resistance towards switching currently established batch and fed-batch manufacturing to novel continuous technologies. Regulators also expect manufacturers to develop tailored control strategies to monitor their processes and ensure product quality. The sector still has a long way to go before developing truly end-to-end control strategies,” explained Kontoravdi.

“Model-based control solutions are available for a variety of fed-batch upstream bioprocesses; however, these are highly process specific, and the market currently lacks universal and easy-to-use model-predictive control (MPC) options for advanced bioprocess control for an integration of upstream and downstream processes.”

These issues are likely to encourage the adoption of automation and artificial intelligence driven technologies, noted Sarkis.

“Robust process performance can increase process efficiency and support manufacturers in meeting quality, economic and environmental objectives. In this space, automated bioprocessing provides numerous advantages, including improved process control, enhanced product quality, reduced variability, increased efficiency, and cost savings,” she said.

The post Tech is Key to Sustainable Biopharma appeared first on GEN - Genetic Engineering and Biotechnology News.

According to Papantoniou, who spoke at the Terrapinn Advanced Therapies Congress earlier this year, the huge quantities of data required to train machine learning algorithms mean it’s essential to already have necessary sensor technology, automation hardware, and process control and analytics in place.

“You need to implement automation in a systematic way or else it’s just numbers—like a brain with no hands,” he explains.

Papantoniou explains that among his team’s projects, they used machine learning to study the growth kinetics of donor progenitor cells in scale-down suspension bioreactor systems. However, to acquire meaningful data, they trained the AI on data collected on cells from more than two hundred donors.

Meaningful AI

“To get meaningful AI, you need more donor cells than are available to most people developing cell-based products,” he points out. “The AI has to be trained with a given number of data points, calibrated, and validated with another, independent control dataset.”

At that time, he felt the advanced therapy industry was producing enough data to benefit from AI. But now, he says, that’s beginning to change with the adoption and integration of new automated technologies. These include smart bioreactors from companies, such as Ori Biotech or a variety of biosensors, such as Raman Spectroscopy probes.

“People are now making multiplex sensors, which can be cheaply incorporated into disposables used for cell differentiation and expansion monitoring. This expands the range of critical quality attributes you can measure,” points out Papantoniou, adding that he expects this trend to continue, making it more feasible to use AI for predicting complex cell behaviour while controlling manufacturing processes.

Papantoniou’s team is currently using a Horizon Europe grant to build a small factory to manufacture organoids, organ-like tissues that can be used for the bioprinting of larger tissues. The factory, which includes bioreactors, bioprinting, high-speed imaging, and robotic liquid handling, is designed to monitor proteins and metabolites secreted by the cells, allowing 3D tissue implants to be carefully controlled by AI.

“We know what to measure to support cellular and tissue development and we’re now growing more confident in using these data to predict how cells might behave,” he says.

Further into the future, Papantoniou predicts that AI may even be used for managing decentralized manufacturing at sites based in different countries.

The post Advanced Therapy Needs Well-Controlled Processes to Benefit from AI appeared first on GEN - Genetic Engineering and Biotechnology News.

“This technology pipeline really enables us to analyze the human brain at multiple scales. Potentially this pipeline can be used for fully mapping human brains,” said study senior author Kwanghun Chung, PhD, an investigator at Picower Institute for Learning and Memory at MIT.

Chung and his team published their work, titled, “Integrated platform for multiscale molecular imaging and phenotyping of the human brain,” in the latest issue of Science.

The process of sectioning, staining, and photographing tissue samples, especially large samples, like the whole human brain, is typically destructive, limiting the data that can be gathered from each sample. In the new report, Chung’s team performed “holistic imaging of human brain tissues at multiple resolutions from single synapses to whole brain hemispheres and we have made that data available,” the senior investigator said.

To accomplish this task, members of Chung’s team created a suite of three key innovations. First, Ji Wang, PhD, developed the Megatome, an improved vibratome, which slices intact human brain hemispheres finely without causing damage. The vibratome’s precision in blade movement and sample stability not only allows for the sample to remain undamaged at cut sites, but also cuts at a higher speed than similar vibratomes. These improvements allow for thick tissue slabs to be prepared much more quickly and precisely.

Although the tissue slabs are thicker than usual for research purposes, a specially designed hydrogel technology has been developed by Juhyuk Park, PhD, for imaging purposes. The hydrogel, mELAST, is used to infuse the brain sample prior to sectioning. The samples, and slices, are clear, flexible, durable, and expandable, retaining tissue integrity and allowing for repeated use in future studies.

Finally, once the slabs are imaged, those visual data can be stitched together into a full three-dimensional organ reconstruction using a computational system called UNSLICE, which has been developed by Webster Guan, PhD. UNSLICE correctly aligns and combines the sample images, with its algorithms that trace anatomical features, like blood vessels or axons, between slices to match them correctly and precisely.

“We need to be able to see all these different functional components—cells, their morphology and their connectivity, subcellular architectures, and their individual synaptic connections—ideally within the same brain, considering the high individual variabilities in the human brain and considering the precious nature of human brain samples,” Chung said. “This technology pipeline really enables us to extract all these important features from the same brain in a fully integrated manner.”

This technology can be used by researchers to examine more quickly and efficiently brain tissue, while preserving the tissue for multiple studies and additional experiments. As a proof-of-concept, Chung’s team collaborated with co-author Matthew Frosch, MD, a physician at Massachusetts General Hospital, to explore the use of this technology in Alzheimer’s disease research.

The joint venture included a full examination and comparison of two brain hemispheres, one with Alzheimer’s and a control. “We didn’t lay out all these experiments in advance. We just started by saying, ‘OK, let’s image this slab and see what we see,’” Chung said. Their goal was to determine if the technology they developed could improve visualization and understanding of brain anatomy and microanatomy in a diseased brain. Chung added, “We identified brain regions with substantial neuronal loss so let’s see what’s happening there. ‘Let’s dive deeper.’ We used many different markers to characterize and see the relationships between pathogenic factors and different cell types.”

This study showed both the functionality and image quality of the imaging technology along with its utility and potential for comparative and exploratory studies. The technology’s utility appears to extend beyond the brain to other organs and tissues. In their paper, the authors concluded: “We envision that this scalable technology platform will advance our understanding of the human organ functions and disease mechanisms to spur development of new therapies.”

The post 3D Brain Imaging Tool Provides Holistic Analysis Down to Subcellular Features appeared first on GEN - Genetic Engineering and Biotechnology News.

The publication, which is a collaboration between DeepMind and Isomorphic Labs, a London-based company launched in 2021 to build upon DeepMind’s AI research to tackle biological problems, emphasizes AlphaFold 3’s potential to advance therapeutic applications, including drug discovery.  

Sir Demis Hassabis, PhD, co-founder and CEO of DeepMind and Isomorphic Labs, declared he was “thrilled to announce AlphaFold 3, which can predict the structures and interactions of nearly all of life’s molecules with state-of-the-art accuracy… Biology is a complex dynamical system so modeling interactions is crucial.” 

But his enthusiasm was not universally shared. In contrast to the launch of AlphaFold 2 in 2021, Nature’s publication of AlphaFold 3 lacked the open source code. That omission has sparked outcry from the research community, culminating in a protest letter signed by more than 1,000 scientists.

Within an hour of AlphaFold 3’s publication, Stephanie Wankowicz, PhD, a computational biologist at the University of California, San Francisco, says there were “emails flying” between structural biologists, chemical biologists, and more exclaiming, “how can they make claims like this? We can’t test them. We can’t reproduce them.”  

In lieu of code, AlphaFold 3 was released to the community as a web server, accessible to individuals without a computational background. While a recent blog post from DeepMind and Isomorphic Labs described the new AlphaFold Server as a facilitator of “novel hypotheses to test in the lab, speeding up workflows and enabling further innovation,” researchers were quick to note the web server’s limitations. 

Scientists took to social media to complain how ligand inputs to the web server were limited to a finite list of biological molecules and did not allow custom inputs, such as small molecule drugs. In that vein, while Isomorphic Labs has reported applying AlphaFold 3 to drug design for internal projects as well as with pharmaceutical partners, the public web server is under a license that is limited to non-commercial use. 

Upon launch, the AlphaFold Server also imposed a daily limit of 10 requests per day. While that number later increased to 20, with DeepMind planning to “explore other approaches for quota allocation in the future, including weekly or monthly allocations,” high-throughput studies currently remain inaccessible. 

“If you’re a cell biologist interested in a particular protein-protein interaction, you could manually input a few jobs [into the AlphaFold 3 web server] and get a result for a couple of systems,” said Roland Dunbrack, PhD, professor at the Fox Chase Cancer Center in Philadelphia, who was one of the original reviewers of the AlphaFold 3 Nature paper.  

“But if you’re in structural bioinformatics where you’re interested in methods development or technology, then you really need to be able to do high-throughput calculations,” Dunbrack continued. 

Dunbrack emphasizes that these access restrictions prevent the scientific community from applying AlphaFold 3 to its full potential. In his review of the original AlphaFold 3 manuscript, Dunbrack described how AlphaFold 2’s open source code allowed other researchers to make further advances such as the establishment of bespoke servers for ease of use, high-throughput structure prediction, and method development for specific protein types or structural features. These applications were not made directly available by DeepMind. 

“It was an incredible disservice to science that [DeepMind] did this wonderful thing with AlphaFold 2, but the lack of availability and the things that we could possibly do with [AlphaFold 3] was just a big step backwards,” Dunbrack told GEN Biotechnology. 

Tellingly, after insisting to the Nature editors that “publishing the paper without code would be a mistake,” Dunbrack was not given the opportunity to review the revision prior to publication despite multiple outreach attempts to Nature. 

Mohammed AlQuraishi, PhD, assistant professor in the department of systems biology at Columbia University, notes that not being able to access the model weights presents additional restrictions for research building on AlphaFold 3.  

“After AlphaFold 2 came out, many people spent time analyzing the behavior of the system and understanding its failure and success modes. This line of research has been incredibly valuable in elucidating how and what these systems learn. This will not be possible for AlphaFold 3 until DeepMind releases the model weights or someone reproduces their results,” AlQuraishi told GEN Biotechnology.  

In a Nature Methods paper, published a few days after AlphaFold 3, AlQuraishi and Nazim Bouatta, PhD, senior research fellow at Harvard Medical School, presented OpenFold, a fast, memory efficient and trainable implementation of AlphaFold 2. According to AlQuraishi, OpenFold was used to better understand how AlphaFold generalizes to unseen regions of structure space and found that it is quite robust. For example, the system could still predict the rough geometry of beta sheets even if it was only trained on alpha helices. 

AlQuraishi added that independently reproducing AlphaFold 3 would be highly valuable to permit training new variants of the system, including potentially training on private data repositories. To pursue this goal, his group has begun to reproduce AlphaFold 3 based on the Nature’s publication of pseudocode, or the representation of code used to describe the implementation of an algorithm.  

Deviation from community standards 

Among the social media outcry was an open letter submitted to Nature’s editors, co-authored by researchers in the structural and computational biology field, including Wankowicz and Dunbrack.  

The letter amassed more than 1,000 signatures by the end of May and outlined “several deviations from our community’s standards,” including not making the code available to peer reviewers despite “repeated requests,” which the letter authors describe as a failure by the journal to enforce its own policies.  

“I have no doubt that [the ability to predict molecular interactions with AlphaFold 3] could be a transformative advance for the field,” Wankowicz told GEN Biotechnology. “That’s why this evokes such a strong response from the scientific community. We want access to it because we know that this is probably really good science.” 

Nature’s editor-in-chief, Magdalena Skipper, PhD stated that when making a decision on data and code availability, the journal reflects on several factors including “the potential implications for biosecurity and the ethical challenges this presents.” In such cases, the journal works with authors to provide alternatives that will support reproducibility, such as pseudocode.   

However, John Jumper, PhD, one of the senior authors of the AlphaFold 3 article and lead author on AlphaFold 2, reportedly told the press that DeepMind and Isomorphic Labs had consulted more than 50 experts in biosecurity, bioethics, and AI safety and concluded that AlphaFold 3’s marginal biosecurity risks were far outweighed by the system’s potential benefits to science.

Max Jaderberg, PhD, chief AI officer at Isomorphic Labs, and Pushmeet Kohli, PhD, vice president of research at DeepMind, stated that the team is “working on releasing the AF3 model (incl weights) for academic use” within six months. Nature said the journal would update the published paper with the code once it is released. When asked for further comment, DeepMind directed inquiries to the AlphaFold Server FAQ section. 

Many researchers are also dissatisfied with Nature’s response in its latest editorial, which described the decision not to publish the code as an “opportunity for conversation” at a time when the majority of global research is privately funded and not published in peer-reviewed journals. “We at Nature think it’s important that journals engage with the private sector and work with its scientists so they can submit their research for peer review and publication,” the editorial continued. 

But AlQuraishi responds that the issue is a “seemingly double standard that Nature applies,” where academics are expected to provide the source code behind their work but the same is not true of an industrial lab. “The concern here, of course, is that this may lower the openness standards in the field,” said AlQuraishi.  

“[Nature] seems to have set a two-tiered set of standards, one for academics, one for for-profits. This is an incredibly poor precedent to set. Peer-reviewed research should be held to the same standard, regardless of the author’s name or affiliations,” weighed in Wankowicz. 

“Many companies want the Nature ‘stamp’ of approval,” said James Fraser, PhD, chair of the department of bioengineering and therapeutic sciences at UCSF. “This editorial shows, nakedly, that this ‘stamp’ is a toxic part of our current research ecosystem, one that bends easily to corporate interests and applies inequitable standards.” 

In a letter sent to Nature and posted on the social media platform, X, Anshul Kundaje, PhD, associate professor of genetics and computer science at Stanford University, wrote that while commercial entities are under no obligation to open source or share details about their products, “this does not mean they get to bypass canonical standards for what constitutes a peer-reviewed and verifiable scientific publication. What Nature published as a peer-reviewed article is in fact an advertisement and at best a white paper.”  

Taken together, the publication of AlphaFold 3 has sparked a larger conversation within the scientific community around the changing landscape of research tools and the role of journals in communicating science to ensure reproducibility and accessibility to allow further scientific progress. That conversation will continue as the field eagerly awaits the release of AlphaFold 3’s nuts and bolts. 

A note of caution, however, was injected by Derek Lowe, PhD, medicinal chemist and veteran blogger at In the Pipeline. “Structure is not everything,” Lowe wrote shortly after the Nature report was posted. “It’s very useful, very good to have, and it will accelerate a lot of really useful research. But it does not take you directly to a drug, nor to a better idea about a target for a drug, nor to a better chance of passing toxicity tests, nor to a better chance of surviving oral dosing and the bloodstream and the liver. Better structure predictions are tools that we can use to attack those crucial problems, but they don’t answer any of them. Drug discovery has not been solved by software, no matter what you might read.”

This article was published in the June 2024 issue of GEN‘s sister peer review journal, GEN Biotechnology.

Fay Lin, PhD, is senior editor for GEN Biotechnology.

The post AlphaFold 3 Angst: Limited Accessibility Stirs Outcry from Researchers appeared first on GEN - Genetic Engineering and Biotechnology News.