I'll describe the main elements of our AI-driven experimental platform (including models like NeuralPLexer for structure prediction, and plate-based experimentation for closed-loop exploration and data generation). As an example of how these technologies have been effectively brought to bear in discovery projects at Iambic, I'll talk through our lead program, which went from launch to IND in 24 months, and is now being studied in a Phase 1 clinical trial.

This presentation will introduce the DIAGONAL platform, a cutting-edge integration of ML and physics-based methodologies that overcome technical limitations that hindered agonist antibody discovery in the past and allows for reproducible and reliable design of developable agonist antibodies against any multimeric receptor complex. We will illustrate the method by presenting the four case studies where the DIAGONAL platform helped us to design antibody agonists targeting four structurally distinct receptor complexes with a 100% success rate.

Kinase inhibitors are a successful category of therapeutics used in treating a wide variety of disease. Despite their efficacy, these drugs often present clinically relevant adverse events that can limit their therapeutic utility. To improve modeling of kinase inhibitor toxicity in patients, we extensively mined both literature and publicly available toxicity data to develop the Kinase Anti-Targets Compendium, a database of kinases statistically associated with specific toxicities.

Navigating through chemical space to find potent, selective, ADMET-compliant molecules for novel targets is challenging, since data is available for very few ligands. We demonstrate that with only 500 potency measurements, active learning on a foundation model can find better molecules than a 1M molecule high-throughput screen. Our approach searches over all synthesizable, drug-like molecules more effectively that state-of-the-art algorithms like REINVENT and Graph GA. This remarkable efficiency is enabled by our foundation model's accurate potency predictions given minimal training data.

We present Exscientia's approach to drug discovery and the important role of generative small molecule design in the process. The discovery of a novel LSD1 inhibitor and a successful proof-of-concept study leading to the discovery of novel, bispecific anti-malaria agents are presented as applications of generative design.

We have developed new antibacterial small molecules by targeting an RNA hairpin within the ribosomal PTC, using a unique combination of NMR and computational chemistry. The optimization models utilized a dataset of small molecules, and their docking scores were essential in establishing design principles for new small molecules with improved bioactivity. We then synthesized these molecules, tested their ability to inhibit the ribosome, and demonstrated their binding mechanism to the RNA hairpin.

Unlike traditional AI that relies on existing knowledge, Anima's Lightning AI, a pioneering Visual Biology platform for drug discovery, generates new disease data by imaging cellular pathways through large-scale experimental biology. It trains neural networks on millions of visualizations, enabling them to visually recognize disease mechanisms, identify dysregulated pathways and novel, experimentally validated targets. It offers unbiased high-content, high- throughput screening of small molecules that revert disease signatures to their healthy state by modulating the mRNA biology of hard targets. 

Early translational studies using human-relevant in vitro models can help our understanding of disease pathobiology by simulating hallmark phenotypes in patients. To study these phenotypes, we have developed both 2D and 3D in vitro human models using iPSC-derived cardiomyocytes. We have analyzed the behavior of these models using computational methods like signal processing and computer vision. We will present promising results from this analysis as well as potential caveats.

This presentation will showcase examples from FDA projects in Generative AI, focusing on its application using Generative Adversarial Networks (GANs) in predictive toxicology, translational toxicology, and drug safety assessment. Specifically, these projects apply GANs to: (1) Learn from existing animal study data to generate animal study results without conducting actual experiments; (2) Generate toxicogenomics data based solely on experimental design; and (3) Translate findings from one organ to another. Additionally, the presentation will include a comparative analysis of discriminative and generative approaches, highlighting their complementary roles in toxicology and drug safety. Finally, a framework will be discussed to integrate both discriminative and generative AI to advance the 3Rs (Replacement, Reduction, and Refinement) in toxicology, drug discovery, and safety assessment.

Spatial profiling technologies like hyper-plex immunostaining in tissue, spatial transcriptomics, coupled with non-spatial profiling like scRNAseq have the potential to enable a multi-factorial, multi-modal characterization of the tissue microenvironment. Objective scoring methods inspired by recent advances in statistics and machine learning can serve to aid the interpretation of these datasets, as well as their integration with companion data like bulk and single cell genomics. I will discuss analysis paradigms from machine learning that can be used to integrate and then prioritize gene regulatory programs underlying oncogenesis (as well as therapeutic candidates) using case studies.

Efficient syntheses of complex small molecules often involve speculative experimental approaches. The central challenge of such plans is that experimental evaluation of high-risk strategies is resource intensive, as it entails iterative attempts at unsuccessful strategies. This presentation describes a complementary strategy that combines creative human-generated synthetic plans with robust computational prediction of synthetic feasibility using computational modeling, density functional theory, and machine learning  This work defines how machine learning models can drive complex molecule synthesis.