AI is impacting drug discovery through agentic workflows and advances in co-folding such as OpenFold3. Yet predicting molecular behavior and properties remains challenging. We explore how to build foundation models with quantum mechanics, molecular dynamics, and experimental feedback to improve predictions of affinity, energetics, properties, and function. Drawing on PsiThera's experience, we discuss how AI is advancing therapeutic discovery while empowering drug designers with direct access to sophisticated computational tools.

ArrePath's oral bioavailability model, trained only on public data, predicts F on series it has never seen. It guided an antibacterial program from ~10% to >80% F, surfacing non-obvious orally bioavailable compounds. Our AI agent runs this model alongside potency and ADME models, so one scientist can evaluate 10,000+ designs a week and prioritize bioavailable compounds from the first design cycle. We present case studies from real drug discovery campaigns.

Recent clinical successes have reawakened interest in cyclic peptides. However, simultaneous design of permeability and potency remains a significant challenge. We present a multi-technology approach combining physics-based methods and machine learning to perform multiparameter design, generating cyclic peptides with robust PAMPA permeability and nanomolar potency. Critically, unlike ML-only methods—which are biased by their training sets and largely limited to the 20 canonical amino acids—our approach explores an unbiased chemical space.

We present an AI-driven approach to de novo macrocyclic peptide design directly from target structures, aiming to achieve picomolar binding affinity and sufficient metabolic stability while enabling therapeutically relevant exposure even with limited intestinal absorption. Using a validated cytokine receptor as a case study, we demonstrate the progression from binding interface identification to experimentally characterized macrocyclic peptide leads, highlighting the balance between potency, stability, and intestinal absorption.