Foundation Models for AI-Enabled Biological Design
Asher Moldwin, Amarda Shehu
TL;DR
This paper addresses how foundation models can enable AI-enabled biological design by surveying architecture families (Transformer, diffusion, and state-space models) and their applications to protein, small-molecule, and genomic sequence design. It provides a taxonomy of architectures, controllability methods (fine-tuning, conditional generation, RL, and post-generation filtering), and multi-modal integrations, highlighting concrete models and design objectives across CLMs, PLMs, and GLMs. Key contributions include a structured overview of model families, generation-control strategies, and open problems with actionable directions for future work, such as domain-specific self-supervised objectives and standardized benchmarks. The work underlines the high practical impact of FM-enabled design for drug discovery, synthetic biology, and genetic engineering, while stressing the need for standardization and rigorous evaluation as the field matures.
Abstract
This paper surveys foundation models for AI-enabled biological design, focusing on recent developments in applying large-scale, self-supervised models to tasks such as protein engineering, small molecule design, and genomic sequence design. Though this domain is evolving rapidly, this survey presents and discusses a taxonomy of current models and methods. The focus is on challenges and solutions in adapting these models for biological applications, including biological sequence modeling architectures, controllability in generation, and multi-modal integration. The survey concludes with a discussion of open problems and future directions, offering concrete next-steps to improve the quality of biological sequence generation.