This presentation explores how AI and data platforms can enhance the landscape of research and development, with an ultimate goal to reduce timelines, increase probability of success, and better predict which patients are most likely to benefit from our medicines. We will describe the robust data foundation we have built and the prerequisites for deploying AI at scale.

This talk focuses on optimizing the logistical processes of preclinical animal studies through technology, including aspects of improved data management, tracking systems, or automation in animal handling, all aimed at speeding up the early stages of drug development.

This presentation will explore how Causal AI and mechanistic patient stratification enhance R&D success in precision medicine, focusing on complex, chronic diseases. Attendees will learn how these technologies drive the discovery of personalized treatments and innovative diagnostics for unmet medical needs. We’ll discuss strategies for identifying novel targets aligned with patient biology, designing faster, more successful clinical trials, and connecting patients to effective therapies based on their disease mechanisms.

Drug repurposing offers a promising avenue for rapidly identifying existing therapeutics for new indications, reducing the time and cost associated with traditional drug development. This study presents a robust multimodal pipeline designed to streamline the drug repurposing process. By integrating diverse computational techniques, we constructed and analyzed extensive networks from datasets encompassing 4,136 drug targets and 24,554 drug interactions. Utilizing tools such as Cell Designer, Cytoscape, and DrugX, we mapped complex relationships between drugs, their targets, and associated interactions, providing a comprehensive overview of potential repurposing candidates. Our platform, comprising 11 specialized modules, systematically employs specific datasets, algorithms, and scoring systems to identify, evaluate, and rank potential drug candidates. The networks generated—such as Drug-CoV, Drug-Drug, Drug-Side-Effect, and Drug-Human—were critical in predicting drug properties and understanding their interactions within biological systems. The DrugX tool, specifically, enabled the detailed assessment of drug side effects and their similarity to target disease symptoms, further refining candidate selection. This platform not only enhances the efficiency of drug repurposing efforts but also provides a scalable framework that can be applied across various therapeutic areas. By leveraging artificial intelligence and network-based analysis, our approach accelerates the identification of viable drug candidates, facilitating faster transitions from discovery to clinical application.

Tesselogic is a new active learning framework to accelerate target discovery in any disease model, from cancer cell lines to complex, organotypic cultures of patient-derived cells. We can identify top targets with as little as 3% of the cost and effort of genome-wide screening. We have applied our approach to target discovery for lung disease in house and benchmarked it on >1,000 screens from the BioGRID ORCS database across 31 tissues of origin.

The SynTech group at Novartis focuses on leveraging synthesis and technology to impact drug discovery. This talk will illustrate our application of automation and machine learning to accelerate the design-make-test-analyze cycle. Particular emphasis will be placed on how we utilize models and active learning techniques in molecular design, as well as our efforts to democratize these tools within the medicinal chemistry community.

A key challenge in drug discovery is designing compounds that bind with high affinity to a target protein’s pocket. A wide range of computational methods has been developed to assess binding affinity. These range from fast docking algorithms to more rigorous, physics-based approaches. In this talk, we present how we use physics-based absolute binding free energy calculations during hit identification and optimization phases to facilitate the prediction of protein-ligand binding affinities, guiding drug design. Machine learning predictions are also performed to predict small molecules properties. Moreover, we provide the results of absolute binding free energy calculations which we performed on a wide range of protein-ligand complexes. Finally, we showcase our MD-based approach to characterizing novel protein-protein interfaces, highlighting its impact along with internal AI models on small molecule identification and optimization. 

This presentation examines how merging quantum computing with AI can revolutionize drug discovery and biomarker mapping in the biopharmaceutical industry. By leveraging quantum computing's immense computational, AI's predictive capabilities, and a new biological framework developed by FindInfinite Labs, we can overcome current limitations in biological computations and accelerate the development of new therapies while at the same time use new predictive and quantum AI models to accelerate our understanding of the complex nature of life.

After “talking the talk,” AI-centric drug discovery companies (AIDDCs) are now “walking the walk” by not only progressing lead molecules to clinical trials but also disclosing novel chemical series for drug targets in published patents. This presentation outlines the mining of these documents using open sources to provide a valuable snapshot of what data AIDDCs are actually generating. Example filings include WO2022106857 from Exscientia (targeting MALT1), WO2020039209 from BenevolentAI (targeting TRK), WO2021155253 from Atomwise (targeting ANAT), and WO2021219089 from Insilico Medicine (targeting Cov2 M-protease). This presentation will demonstrate how open extraction resources were used to map selected AI-generated compounds into PubChem, exploring the novelty of their chemical and target neighborhoods.

Large-scale information extraction from patent publications in the pharma and biotech domain is fundamental for early R&D, yet remains a challenging and expensive process due to the inherent complexity of IP documents. We have developed the PatentMiner—an end-to-end AI-powered tool for automated table and free-text extraction from patents. It is designed as a modular Python package which combines state-of-the-art OCR algorithms coupled with uncertainty estimates, multimodal LLM-based as well as rule-based error correction, and automated post-processing. In addition, an intuitive user interface allows for visual inspection and manual error correction if necessary. The PatentMiner is used in several early oligonucleotide-related projects, thereby significantly reducing the time spent to accurately extract both structured and unstructured data from relevant patents.

Envisagenics' SpliceCore platform leverages advanced AI to accelerate immunotherapy drug discovery by identifying novel, tumor-specific epitopes arising through RNA splicing. This innovative approach focuses on splicing events that drive tumor progression and generate new therapeutic targets. Supported by comprehensive case studies and experimental validation, SpliceCore demonstrates robust predictive capabilities, offering a powerful tool for discovering safe, highly specific epitopes. SpliceCore represents a significant advancement in the development of next-generation immunotherapies.

The majority of human genes do not encode proteins and are not conserved in evolution beyond mammals. Nevertheless, drug discovery has historically focused on conserved, protein-coding targets and their pathways. This talk presents cutting-edge research on RNA-based drugs targeting an evolutionarily young long non-coding RNA (lncRNA) identified through Genome-Wide Association Studies. These innovative therapeutics demonstrate early potential to replace insulin and GLP-1 receptor agonists in managing diabetes within a nonhuman primate model. The presentation highlights the discovery process and translational significance of lncRNA-targeted drugs, heralding a novel approach to treating metabolic disorders and advancing precision medicine.