We describe a novel method that combines experimental biophysical data (HDX-MS) with weighted ensemble simulations (WES) to accurately predict ternary complex structures at atomic resolution. We show that the WES+HDX approach generates accurate structures (RMSD below 2.0 Å to x-ray) and can reproduce solution-state dynamic behavior of the ternary complex. We illustrate how this approach is used to predict degradation propensity of different heterobifunctional and glue molecules.

Chemo-proteomics profiling of PROTACs designed from pan-class inhibitors revealed a large difference in the degradation frequencies of the target proteins engaged by these molecules. Using protein-intrinsic features, we developed a machine learning classifier that discriminates target proteins based on their observed degradation patterns and highlights properties that dictate their degradability. Using computational structural modeling, we provide mechanistic insight into the predicted features and obtain actionable information for rational PROTAC design.

Combining the state-of-the-art molecular modeling, in heterogeneous data sources and in machine learning techniques, we are dramatically increasing the accuracy in our computational predictions. We will showcase recent successful case studies including virtual screening enrichment, ligase screening for TPD and ternary complex formation in PROTACs. Overall, the enrichment of machine learning techniques with data augmentation from molecular modeling seems to provide the necessary boost that prediction models might need.

Drug discovery has benefitted from the application of AI to enhance the ability to deal with large, diverse data coming from genomics, imaging, EHR'S and claims data. While this has greatly improved operational efficiency, there remains a significant gap between drug discovery and actual management of disease. An introduction to Next-Generation Phenotyping (NGP) in multiple sclerosis will be presented to highlight the challenges and opportunities.

Given the wide variety of AI tools currently being deployed in radiology and pathology, it is feasible to bring AI to pre-clinical evaluations for drug trials. However, due to its direct impact on response evaluation and trial success metrics, its important to understand their calibration properties and quality. Using the example of tumor segmentation task, we will review a few algorithms and describe their potential modes of failure in addition to scoring rubrics related to data and model veracity. Then, using case studies from Pathology, we will then examine "failure modes" in the performance of AI algorithms in that space. 

Immunotherapies have shown limited efficacy in treating glioblastoma. While radiographic tumor contact with the lateral ventricle correlates with worse outcomes; the extent to which ventricle proximity impacts immunobiology in the tumor microenvironment remains unknown. Using CyTOF profiling and machine learning approaches, we identify the suppressive impact of ventricle contact on anti-tumor immunity in the brain and reveal potential clinically actionable immune targets and patient stratification methods for glioblastoma.

NCATS through the ASPIRE (A Specialized Platform for Innovative Research Exploration) program seeks to transform the design-synthesize-test cycle through the development of new algorithms and workflows to capture data from automated chemical synthesis and biological testing systems to predict and inform the next iteration of new chemical entities. ASPIRE will enhance the ability to discover and develop new chemistries towards previously undrugged biological targets (i.e. those with no known drugs to modulate their function) across many human diseases and conditions.

In recent years, we have seen significant advancements in the field of precision medicine, primarily in oncology settings where druggable driver mutations are present. New genomic technologies hold great promise for the identification of actionable drug targets and associated biomarkers for several complex diseases, such as autoimmunity. However, the discovery of novel targets in these settings is often complicated by multigenic effects and the involvement of multiple cell types in disease progression. Our approach uses single-cell RNAseq and machine learning to elucidate the precise cell types involved in the progression of complex diseases and to identify novel therapeutic targets.

In this talk, I will describe our continued efforts to use genomics and AI to identify the targets of compounds that may not have entirely known mechanisms of action. I will describe how these approaches can be used to screen libraries of compounds in silico to uncover repositioning opportunities. I will then describe the successful application to an anticancer compound, followed by precise clinical positioning in pediatric brain cancers.

Recursion is a clinical-stage pharmatech company, mapping human biology at scale to bring better medicines to patients. Enabled by 14 petabytes of imaging and other omics data, we use deep learning to build biological representations across multiple cell types, a whole-genome CRISPR library, and nearly 1 million compounds. Here, we demonstrate the capability of this platform to model complex disease and identify and optimize compounds as potential cancer therapies. Integrating recent insights on genetic correlates of drug response, we demonstrate the application of inferential search to uncover novel mechanisms and small molecules that enhance response in immunotherapy and HRD- cancers.