We have tested a small set of PROTACs against several human drug transporters in vitro as well as in zebrafish for their effects on embryo and larval mortality, morphology, embryo photomotor response, and larval photomotor response. We will report the preliminary outcomes of these combined assessments and discuss the implications for PROTACs development as well as how this data could be used for machine learning model generation and evaluation.

A major obstacle to cyclic peptide development is that little structural information is available, as most cyclic peptides adopt multiple conformations in solution. By combining molecular dynamics simulation and machine learning, we can now provide simulation-quality cyclic peptide structure predictions in seconds to enable structure-based design of cyclic peptides and an understanding of their sequence–activity relationships.

AI-based modeling and physics-based simulations excel in specific drug discovery areas but often fail in broader molecular and cellular biology applications. Addressing these gaps requires new algorithms. I will discuss AWSEM, a coarse-grained protein simulation method derived from early neural network applications in protein structure modeling. AWSEM excels in predicting structures of macromolecular complexes and simulating their dynamics. Recent work showcases its superior performance in modeling PROTACs, significantly outperforming AlphaFold2. Additionally, I will introduce our novel docking and virtual screening algorithms that integrate physics and AI to achieve state-of-the-art results.

To build better models of small molecule binders to target proteins, high-quality training sets are essential. Leveraging DNA-encoded libraries (DELs) with billions of molecules, we utilize our selection technology to generate data with true negatives and accurately rank ordered true positive binders. We are working to utilize these large-scale data sets to enhance the predictive power of our small molecule protein binding models.

The gap between drug discovery and information science continues to close, hence there has never been a better time to leverage the power of AI/ML techniques to advance our understanding of the relationships between chemical and biological space. NCATS has identified, through the input of the greater scientific community, focus areas that need to be addressed in order to transform the design-synthesize-test cycle to transition to be more data-driven. The ASPIRE Program was created to support the development of AI/ML tools to process captured data to inform the next iteration of the process.

CACHE is a series of prospective benchmarking challenges where compounds predicted by 25 computational teams to bind a pre-defined protein target are procured and tested experimentally and all data shared publicly. Lessons learned from the first two CACHE challenges will be presented.

In the fight against rare diseases, AI technologies offer unprecedented opportunities to accelerate discovery and enhance patient care. We utilize AI to bridge significant gaps in rare disease research, focusing on improving diagnostic accuracy, speeding up drug discovery, and tailoring treatments to individual patient needs. Our approach leverages deep learning and genetic algorithms to identify novel drug targets and predict therapeutic outcomes, thus significantly reducing the time and cost associated with bringing treatments to market. This talk will explore how integrating AI into biotech can transform the landscape of rare disease treatment, emphasizing the potential for substantial social impact through improved accessibility and efficacy of therapies.

Our talk explores how data scientists, ML modelers, medicinal chemists, and others have collaborated to accelerate small molecule drug discovery through our closed-loop discovery process. At its core, closed-loop discovery is an integrated, streamlined pipeline that rapidly propagates information through the design-make-test-analyze cycle. We'll illustrate how each expert contributes to the overarching goal and how we've tackled the challenge of enabling effective collaboration among these diverse personas.

We have developed a computational, genetics-based approach which integrates functional data from GWAS, ontologies, and biological networks to predict potential drug targets with evidence for disease association. We obtained a list of target genes associated with ALS and prioritized potential target genes by integrating structural information and cell-type transcriptomics data.

Interaction-based screening and design are promising novel strategies for hit finding and lead optimization. The key information unit in the interaction realm is a thin interface between the interacting ligand and protein. When fed to deep neural networks, interaction signatures may help to infer structure-activity relationships for unliganded proteins by exploiting structural data across ligands and proteins. In this talk, we will give an overview of the approach and describe its successful applications to several challenging targets.

Identification of hot spots on the surface of macromolecules is key to evaluating the druggability of novel targets and the likelihood of finding new chemical entities. Computational hot-spot mapping was performed across a set of drug targets, and a machine learning approach was employed to select the top druggable sites. Within this context, the utility of AI-generated protein models obtained using AlphaFold, including ensembles of structural models, was assessed.