Mark Gerstein

Protein Liquid-Liquid Phase Separation (LLPS) plays an essential role in cellular processes and is known to be associated with various diseases. However, our understanding of this enigmatic phenomena remains limited. In this work, we propose a graph-neural-network(GNN)-based interpretable machine learning approach to study the intricate nature of protein structure-function relationships associated with LLPS. For many protein properties of interest, information relevant to the property is expected to be confined to local domains. For LLPS proteins, the presence of intrinsically disordered regions (IDR)s in the molecule is arguably the most important information; an adaptive GNN model which preferentially shares information within such units and avoids mixing in information from other parts of the molecule may thus enhance the prediction of LLPS proteins. To allow for the accentuation of domain restricted information, we propose a novel graph-based model with the ability to partition each protein graph into task-dependent subgraphs. Such a model is designed not only to achieve better predictive performance but also to be highly interpretable, and thus have the ability to suggest novel biological insights. In addition to achieving state-of-the-art results on the prediction of LLPS proteins from protein structure for both regulator and scaffold proteins, we examine the properties of the graph partitions identified by our model, showing these to be consistent with the annotated IDRs believed to be largely responsible for LLPS. Moreover, our method is designed in a generic way such that it can be applied to other graph-based predictive tasks with …

In protein biophysics, the separation between the functionally important residues (forming the active site or binding surface) and those that create the overall structure (the fold) is a well-established and fundamental concept. Identifying and modifying those functional sites is critical for protein engineering but computationally non-trivial, and requires significant domain knowledge. To automate this process from a data-driven perspective, we propose a disentangled Wasserstein autoencoder with an auxiliary classifier, which isolates the function-related patterns from the rest with theoretical guarantees. This enables one-pass protein sequence editing and improves the understanding of the resulting sequences and editing actions involved. To demonstrate its effectiveness, we apply it to T-cell receptors (TCRs), a well-studied structure-function case. We show that our method can be used to alter the function of TCRs without changing the structural backbone, outperforming several competing methods in generation quality and efficiency, and requiring only 10\% of the running time needed by baseline models. To our knowledge, this is the first approach that utilizes disentangled representations for TCR engineering.

Single-cell genomics is a powerful tool for studying heterogeneous tissues such as the brain. Yet, little is understood about how genetic variants influence cell-level gene expression. Addressing this, we uniformly processed single-nuclei, multi-omics datasets into a resource comprising >2.8M nuclei from the prefrontal cortex across 388 individuals. For 28 cell types, we assessed population-level variation in expression and chromatin across gene families and drug targets. We identified >550K cell-type-specific regulatory elements and >1.4M single-cell expression-quantitative-trait loci, which we used to build cell-type regulatory and cell-to-cell communication networks. These networks manifest cellular changes in aging and neuropsychiatric disorders. We further constructed an integrative model accurately imputing single-cell expression and simulating perturbations; the model prioritized ∼250 disease-risk genes and drug targets with associated cell types.Summary Figure

A catalog of transcription factor (TF) binding sites in the genome is critical for deciphering regulatory relationships. Here we present the culmination of the modERN (model organism Encyclopedia of Regulatory Networks) consortium that systematically assayed TF binding events in vivo in two major model organisms, Drosophila melanogaster (fly) and Caenorhabditis elegans (worm). We describe key features of these datasets, comprising 604 TFs identifying 3.6M sites in the fly and 350 TFs identifying 0.9 M sites in the worm. Applying a machine learning model to these data identifies sets of TFs with a prominent role in promoting 10 target gene expression in specific cell types. TF binding data are available through the ENCODE Data Coordinating Center and at https://epic.gs.washington.edu/modERNresource, which provides access to processed and summary data, as well as widgets to probe cell type-specific TF-target relationships. These data are a rich resource that should fuel investigations into TF function during development.

Intelligent agents powered by large language models (LLMs) have demonstrated substantial promise in autonomously conducting experiments and facilitating scientific discoveries across various disciplines. While their capabilities are promising, they also introduce novel vulnerabilities that demand careful consideration for safety. However, there exists a notable gap in the literature, as there has been no comprehensive exploration of these vulnerabilities. This position paper fills this gap by conducting a thorough examination of vulnerabilities in LLM-based agents within scientific domains, shedding light on potential risks associated with their misuse and emphasizing the need for safety measures. We begin by providing a comprehensive overview of the potential risks inherent to scientific LLM agents, taking into account user intent, the specific scientific domain, and their potential impact on the external environment. Then, we delve into the origins of these vulnerabilities and provide a scoping review of the limited existing works. Based on our analysis, we propose a triadic framework involving human regulation, agent alignment, and an understanding of environmental feedback (agent regulation) to mitigate these identified risks. Furthermore, we highlight the limitations and challenges associated with safeguarding scientific agents and advocate for the development of improved models, robust benchmarks, and comprehensive regulations to address these issues effectively.

Numerous statistical methods have emerged for inferring DNA motifs for transcription factors (TFs) from genomic regions. However, the process of selecting informative regions for motif inference remains understudied. Current approaches select regions with strong ChIP-seq signal for a given TF, assuming that such strong signal primarily results from specific interactions between the TF and its motif. Additionally, these selection approaches do not account for non-target motifs, i.e. motifs of other TFs; they presume the occurrence of these non-target motifs infrequent compared to that of the target motif, and thus assume these have minimal interference with the identification of the target. Leveraging extensive ChIP-seq datasets, we introduced the concept of TF signal ‘crowdedness’, referred to as C-score, for each genomic region. The C-score helps in highlighting TF signals arising from non-specific interactions …

Precision of transcription is critical because transcriptional dysregulation is disease causing. Traditional methods of transcriptional profiling are inadequate to elucidate the full spectrum of the transcriptome, particularly for longer and less abundant mRNAs. SHANK3 is one of the most common autism causative genes. Twenty-four Shank3 mutant animal lines have been developed for autism modeling. However, their preclinical validity has been questioned due to incomplete Shank3 transcript structure. We applied an integrative approach combining cDNA-capture and long-read sequencing to profile the SHANK3 transcriptome in human and mice. We unexpectedly discovered an extremely complex SHANK3 transcriptome. Specific SHANK3 transcripts were altered in Shank3 mutant mice and postmortem brains tissues from individuals with ASD. The enhanced SHANK3 transcriptome significantly improved the detection rate for potential deleterious variants from genomics studies of neuropsychiatric disorders. Our findings suggest the stochastic transcription of genome associated with SHANK family genes.

A major challenge in near-term quantum computing is its application to large real-world datasets due to scarce quantum hardware resources. One approach to enabling tractable quantum models for such datasets involves compressing the original data to manageable dimensions while still representing essential information for downstream analysis. In classical machine learning, variational autoencoders (VAEs) facilitate efficient data compression, representation learning for subsequent tasks, and novel data generation. However, no model has been proposed that exactly captures all of these features for direct application to quantum data on quantum computers. Some existing quantum models for data compression lack regularization of latent representations, thus preventing direct use for generation and control of generalization. Others are hybrid models with only some internal quantum components, impeding direct training on quantum data. To bridge this gap, we present a fully quantum framework, -QVAE, which encompasses all the capabilities of classical VAEs and can be directly applied for both classical and quantum data compression. Our model utilizes regularized mixed states to attain optimal latent representations. It accommodates various divergences for reconstruction and regularization. Furthermore, by accommodating mixed states at every stage, it can utilize the full-data density matrix and allow for a "global" training objective. Doing so, in turn, makes efficient optimization possible and has potential implications for private and federated learning. In addition to exploring the theoretical properties of -QVAE, we demonstrate its …