In this deep dive, BioSpace examines how small, medium and large companies are using artificial intelligence and machine learning to enhance their drug discovery efforts.
In this deep dive, BioSpace examines how small, medium and large companies are using artificial intelligence and machine learning to enhance their drug discovery efforts.
In a poll recently conducted by BioSpace, more than 50 percent of respondents said that on the drug development side, their company could best leverage artificial intelligence and machine learning for target discovery and validation.
Source: BioSpace LinkedIn Poll, Oct. 2023
This was also reflected in the responses of AI/ML leaders from mid- and large-sized biopharma companies who spoke with BioSpace for this special issue of ClinicaSpace. We asked representatives from four companies how they are using AI to enhance their drug development efforts in different use cases.
GSK
One of the most important things for machine learning in drug discovery is actually what to design the medicine against. It doesn’t matter if you have the best medicine in the world, if you have the wrong target, you’re not going to see the clinical effect.
About: An early adopter of AI/ML tools, GSK uses AI to determine the best point at which to intervene in the disease process, for the most patients.
Using AI/ML, GSK has:
- Built models to continuously monitor and quantify the amount of liver fat from patients’ imaging scan data, leading to “a lot more interesting targets” for nonalcoholic steatohepatitis (NASH).
- Built a machine learning algorithm to stratify hepatitis B patients in a Phase IIb study of bepirovirsen in order to identify which subset showed the greatest response; 9 to 10% responded so well they experienced a functional cure.
- Developed large language models on RNA and DNA sequences to predict the effect of genetic variants on mRNA processing, identifying whether the amount of protein is increased or decreased and whether the variant changes the splicing pattern of a particular gene.
Alto Neuroscience
When you point a powerful analytic approach at a lot of data, it’ll find something, so really central to our approach has been the requirement for prospective independent replication of the clinical enrichment effect. That keeps us honest and makes sure the approach doesn’t lead to biases.
About: Launched in October 2021, Alto uses AI-derived brain biomarkers to match patients to its investigational drugs. Alto collected a pair of clinical wins in 2023 with lead asset ALTO-100; first in major depressive disorder, then in post-traumatic stress disorder.
- Alto’s AI-enabled biomarker platform evaluates brain function measures like EEG, computerized behavioral tests and patterns from wearable devices to identify which patients respond clinically to its drug candidates.
- Using EEG, founder Amit Etkin’s Stanford lab found and replicated biomarkers predictive of response to standard-of-care treatments, including antidepressants, brain stimulation and ketamine.
- The same approach has guided the selection of patients in an ongoing Phase IIb study for ALTO-100, which specifically targets major depressive disorder patients with cognition issues.
- After identifying the signal of cognition for ALTO-100, Alto pointed its machine learning algorithms at even more data to see if it was possible to improve on the outcome.
Pfizer
AI and ML models are turbocharging drug discovery and development. It is already assisting our scientists in proposing new therapeutic hypotheses, predicting patient response to therapy and designing large and small molecule therapies.
About: AI is a key component of Pfizer’s drug development efforts, Enoch Huang, vice president, Machine Learning & Computational Sciences, told BioSpace.
- One area where Pfizer is leveraging AI/ML is for viscosity optimization. Viscosity is an important consideration in monoclonal antibody (mAb) development, Huang said, but optimization typically requires multiple production and screening cycles, which slows the therapeutic discovery process.
- A scarcity of training data has prevented prior AI methods from making accurate antibody viscosity predictions in the past. Pfizer is overcoming this limitation with PfAbNet-viscosity (Pfizer Antibody Network for viscosity), a 3D convolutional neural network architecture, to predict the high-concentration viscosity of therapeutic antibodies.
- The application of this method led to a significant shortening of the discovery cycle time for an important project in the Pfizer portfolio, which is now in clinical testing.
Insilico Medicine
Insilico Medicine will be the next great biopharma company driven by AI because its technology is reshaping drug discovery and development.
About: Insilico in August became the first company to bring a drug discovered and developed using generative AI into Phase II clinical trials. Martin Gershon, managing partner and CIO of Endeavor Venture Funds & Venture Studio, which supports and collaborates with the company, called Insilico a “unicorn.”
- Insilico discovered intracellular “Target X” for idiopathic pulmonary fibrosis using AI analysis, and subsequently developed INS018_055, a novel small molecule inhibitor.
- INS018_055 has shown potential in preclinical studies for improved fibrotic disease treatment with fewer side effects.
- Insilico recently announced an exclusive global licensing agreement with Exelixis for ISM3091, an inhibitor of USP1, which has the potential to treat BRCA-mutated tumors. The company landed another big fish in November 2022 when it partnered with Sanofi to develop drug candidates for up to six new targets in a deal worth up to $1.2 billion.
In today’s economy, Gershon said AI startups need to prioritize the development of data, and key to this is partnerships. “Partnerships develop data; data validates commercial viability; commercial viability attracts investors... Insilico Medicine, to me, is the poster child of what to do.”
This article was originally published as a Special Edition of ClinicaSpace, BioSpace‘s weekly newsletter covering biopharma research and clinical trials.
For more insights like the above, subscribe to ClinicaSpace below.
