A Guided Design Framework for the Optimization of Therapeutic-like Antibodies

Amy Wang, Zhe Sang, Samuel Don Stanton, Jennifer L. Hofmann, Saeed Izadi, Eliott Park, Jan Ludwiczak, Matthieu Kirchmeyer, Darcy Davidson, Andrew Maier, Tom Pritsky, Nathan C. Frey, Andrew Martin Watkins, Franziska Seeger

Track: Machine learning: computational method and/or computational results

Nature Biotechnology: Yes

Keywords: clinical, antibodies, antibody, drug, lipinski, biophysics, guided design, optimization, molecular descriptors, developability, proxy assays, benchmark, electrostatics, hydrophobicity

TL;DR: A general sequence-based framework that evaluates and optimizes antibodies for biophysical properties characteristic of clinical-stage antibodies

Abstract: Antibodies must meet stringent developability criteria for successful commercialization—a challenge for machine learning approaches given the limited available data. Selecting candidates with biophysical properties similar to clinical-stage antibodies offers an alternative to data-intensive approaches. However, such methods typically suffer from limited throughput due to structure-based calculations and can eliminate promising candidates through overly strict filtering. By benchmarking classical filtering methods against experimental datasets, with viscosity as a proof-of-concept, we identify an informative set of biophysical definitions (relevant to charge and hydrophobicity). Using these as optimization objectives for guided design, we introduce TherAbDesign, a sequence-based framework that evaluates and optimizes antibodies for developability without requiring structure prediction or physics-based computation. TherAbDesign proposes rational modifications to mimic the properties of successful therapeutic antibodies, which we demonstrate can improve known developability liabilities like high viscosity without explicitly modeling their mechanism of action.

Anonymization: This submission has been anonymized for double-blind review via the removal of identifying information such as names, affiliations, and identifying URLs.

Presenter: ~Amy_Wang2

Format: Yes, the presenting author will definitely attend in person because they attending ICLR for other complementary reasons.

Funding: No, the presenting author of this submission does *not* fall under ICLR’s funding aims, or has sufficient alternate funding.

Submission Number: 106