Search for: All records

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. 

Abstract DNA breathing dynamics—transient base-pair opening and closing due to thermal fluctuations—are vital for processes like transcription, replication, and repair. Traditional models, such as the Extended Peyrard-Bishop-Dauxois (EPBD), provide insights into these dynamics but are computationally limited for long sequences. We presentJAX-EPBD, a high-throughput Langevin molecular dynamics framework leveragingJAXfor GPU-accelerated simulations, achieving up to 30x speedup and superior scalability compared to the original C-based EPBD implementation.JAX-EPBDefficiently captures time-dependent behaviors, including bubble lifetimes and base flipping kinetics, enabling genome-scale analyses. Applying it to transcription factor (TF) binding affinity prediction using SELEX datasets, we observed consistent improvements inR²values when incorporating breathing features with sequence data. Validating on the 77-bp AAV P5 promoter,JAX-EPBDrevealed sequence-specific differences in bubble dynamics correlating with transcriptional activity. These findings establishJAX-EPBDas a powerful and scalable tool for understanding DNA breathing dynamics and their role in gene regulation and transcription factor binding. 

Abstract Simulating DNA breathing dynamics, for instance Extended Peyrard-Bishop-Dauxois (EPBD) model, across the entire human genome using traditional biophysical methods like pyDNA-EPBD is computationally prohibitive due to intensive techniques such as Markov Chain Monte Carlo (MCMC) and Langevin dynamics. To overcome this limitation, we propose a deep surrogate generative model utilizing a conditional Denoising Diffusion Probabilistic Model (DDPM) trained on DNA sequence-EPBD feature pairs. This surrogate model efficiently generates high-fidelity DNA breathing features conditioned on DNA sequences, reducing computational time from months to hours–a speedup of over 1000 times. By integrating these features into the EPBDxDNABERT-2 model, we enhance the accuracy of transcription factor (TF) binding site predictions. Experiments demonstrate that the surrogate-generated features perform comparably to those obtained from the original EPBD framework, validating the model’s efficacy and fidelity. This advancement enables real-time, genome-wide analyses, significantly accelerating genomic research and offering powerful tools for disease understanding and therapeutic development. 

Over the years, many computational methods have been created for the analysis of the impact of single amino acid substitutions resulting from single-nucleotide variants in genome coding regions. Historically, all methods have been supervised and thus limited by the inadequate sizes of experimentally curated data sets and by the lack of a standardized definition of variant effect. The emergence of unsupervised, deep learning (DL)-based methods raised an important question: Canmachines learn the language of life fromthe unannotated protein sequence data well enough to identify significant errors in the protein “sentences”? Our analysis suggests that some unsupervised methods perform as well or better than existing supervised methods. Unsupervised methods are also faster and can, thus, be useful in large-scale variant evaluations. For all other methods, however, their performance varies by both evaluation metrics and by the type of variant effect being predicted.We also note that the evaluation of method performance is still lacking on less-studied, nonhuman proteins where unsupervised methods hold the most promise. 

Abstract MotivationProtein language models based on the transformer architecture are increasingly improving performance on protein prediction tasks, including secondary structure, subcellular localization, and more. Despite being trained only on protein sequences, protein language models appear to implicitly learn protein structure. This paper investigates whether sequence representations learned by protein language models encode structural information and to what extent. ResultsWe address this by evaluating protein language models on remote homology prediction, where identifying remote homologs from sequence information alone requires structural knowledge, especially in the “twilight zone” of very low sequence identity. Through rigorous testing at progressively lower sequence identities, we profile the performance of protein language models ranging from millions to billions of parameters in a zero-shot setting. Our findings indicate that while transformer-based protein language models outperform traditional sequence alignment methods, they still struggle in the twilight zone. This suggests that current protein language models have not sufficiently learned protein structure to address remote homology prediction when sequence signals are weak. Availability and implementationWe believe this opens the way for further research both on remote homology prediction and on the broader goal of learning sequence- and structure-rich representations of protein molecules. All code, data, and models are made publicly available. 

Transformer-based language models have shown promise in genomics but face challenges unique to DNA, such as sequence lengths spanning hundreds of millions of base pairs and subtle long-range dependencies. Although next-token prediction remains the predominant pre-training objective (inherited from NLP), recent research suggests that multi-objective frameworks can better capture complex structure. In this work, we explore whether the Birdie framework, a reinforcement learning-based, mixture-of-objectives pre-training strategy, can similarly benefit genomic foundation models. We compare a slightly modified Birdie approach with a purely autoregressive, next token prediction baseline on standard Nucleotide Transformer benchmarks. Our results show performance gains in the DNA domain, indicating that mixture-of-objectives training could be a promising alternative to next token prediction only pre-training for genomic sequence modeling. 

Recent studies exploring the abilities of transformer-based protein language models have highlighted their performance on the task of remote homology detection, but have not provided datasets or evaluation procedures geared toward properly measuring performance on this task. With the goal of obtaining more informative and reproducible results, we offer a detailed procedure for constructing datasets and evaluating remote homology detection performance in a way that allows detailed analyses to be performed that shed light on the remote homology detection performance throughout the “twilight zone” of low sequence similarity. Using the proposed procedures, we found that three stateof-the-art protein language models exhibit diminishing performance when the pairwise sequence similarity between the query sequence and other proteins is restricted to below 35% identity.