Predict. Explain. Discover.

Abstract

The objective of drug discovery is to accurately infer the effects of treatments on patients. Drug discovery would therefore be greatly improved if there existed computational models that
accurately predicted the response of patients to interventions, since this would allow
practitioners to safely and economically test and optimize a wide range of therapeutic
hypotheses before ever starting a human clinical trial. Even a more “modest” model that could
accurately predict the functional response of a wide variety of cells to genetic and chemical
interventions would be of tremendous value in designing effective and safe therapeutics more
likely to produce positive outcomes in the clinic. Creating such virtual cells has long been a goal
of the computational research community that even today remains an ambition due to the
daunting scale and complexity of the biomolecular interactions mediating cellular function.
Nevertheless, a confluence of technological advances suggest that there has never been a
better time to attempt to build virtual cells. In this perspective, we describe Valence Labs’s vision
for virtual cells as a transformative platform to indus drug discovery. We set the context
for our vision by reviewing historical progress, and outline their integration within a larger
framework of agentic systems that continuously refine our mechanistic understanding of human
physiology. We highlight recent advances in machine learning, computational power, and data
generation that now enable robust simulation of cellular functional responses, and we outline
key modeling considerations, evaluation benchmarks, and a roadmap for future research.