MOTIVATION: Recent pandemics have revealed significant gaps in our understanding of viral pathogenesis, exposing an urgent need for methods to identify and prioritize key host proteins (host factors) as potential targets for antiviral treatments. De novo generation of experimental datasets is limited by their heterogeneity, and for looming future pandemics, may not be feasible due to limitations of experimental approaches. RESULTS: Here, we present TransFactor, a computational framework for predicting and prioritizing candidate host factors using only protein sequence data. It leverages the pre-trained ESM-2 protein language model, fine-tuned on a limited set of experimentally determined host factors aggregated from 33 independent SARS-CoV-2 studies. TransFactor outperforms machine and deep learning baselines and its predictions align with Gene Ontology enrichments of known host factors, but also provide interpretability through a computational alanine scan, enabling the identification of pro-viral protein domains such as COMM, PX, and RRM, that may be used to direct experimental investigations of virus biology and guide rational design of antiviral therapies. Our findings demonstrate the potential of transformer-based models to advance host factor prediction, providing a framework extendable to orthogonal input modalities and other infectious diseases, enhancing our preparedness for current and future viral threats. AVAILABILITY AND IMPLEMENTATION: Source code is available at https://github.com/marsico-lab/TransFactor. A full reproducibility package, including code, trained models, and data, is archived on Zenodo (https://doi.org/10.5281/zenodo.16793684).

RNA-binding proteins (RBPs) are central actors of RNA post-transcriptional regulation. Experiments to profile-binding sites of RBPs in vivo are limited to transcripts expressed in the experimental cell type, creating the need for computational methods to infer missing binding information. While numerous machine-learning based methods have been developed for this task, their use of heterogeneous training and evaluation datasets across different sets of RBPs and CLIP-seq protocols makes a direct comparison of their performance difficult. Here, we compile a set of 37 machine learning (primarily deep learning) methods for in vivo RBP-RNA interaction prediction and systematically benchmark a subset of 11 representative methods across hundreds of CLIP-seq datasets and RBPs. Using homogenized sample pre-processing and two negative-class sample generation strategies, we evaluate methods in terms of predictive performance and assess the impact of neural network architectures and input modalities on model performance. We believe that this study will not only enable researchers to choose the optimal prediction method for their tasks at hand, but also aid method developers in developing novel, high-performing methods by introducing a standardized framework for their evaluation.

BACKGROUND: Although several studies have provided insights into the role of long non-coding RNAs (lncRNAs), the majority of them have unknown function. Recent evidence has shown the importance of both lncRNAs and chromatin interactions in transcriptional regulation. Although network-based methods, mainly exploiting gene-lncRNA co-expression, have been applied to characterize lncRNA of unknown function by means of 'guilt-by-association', no strategy exists so far which identifies mRNA-lncRNA functional modules based on the 3D chromatin interaction graph. RESULTS: To better understand the function of chromatin interactions in the context of lncRNA-mediated gene regulation, we have developed a multi-step graph analysis approach to examine the RNA polymerase II ChIA-PET chromatin interaction network in the K562 human cell line. We have annotated the network with gene and lncRNA coordinates, and chromatin states from the ENCODE project. We used centrality measures, as well as an adaptation of our previously developed Markov State Models (MSM) clustering method, to gain a better understanding of lncRNAs in transcriptional regulation. The novelty of our approach resides in the detection of fuzzy regulatory modules based on network properties and their optimization based on co-expression analysis between genes and gene-lncRNA pairs. This results in our method returning more bona fide regulatory modules than other state-of-the art approaches for clustering on graphs. CONCLUSIONS: Interestingly, we find that lncRNA network hubs tend to be significantly enriched in evolutionary conserved lncRNAs and enhancer-like functions. We validated regulatory functions for well known lncRNAs, such as MALAT1 and the enhancer-like lncRNA FALEC. In addition, by investigating the modular structure of bigger components we mine putative regulatory functions for uncharacterized lncRNAs.

Summary: Convolutional neural networks (CNNs) have been shown to perform exceptionally well in a variety of tasks, including biological sequence classification. Available implementations, however, are usually optimized for a particular task and difficult to reuse. To enable researchers to utilize these networks more easily, we implemented pysster, a Python package for training CNNs on biological sequence data. Sequences are classified by learning sequence and structure motifs and the package offers an automated hyper-parameter optimization procedure and options to visualize learned motifs along with information about their positional and class enrichment. The package runs seamlessly on CPU and GPU and provides a simple interface to train and evaluate a network with a handful lines of code. Using an RNA A-to-I editing dataset and cross-linking immunoprecipitation (CLIP)-seq binding site sequences, we demonstrate that pysster classifies sequences with higher accuracy than previous methods, such as GraphProt or ssHMM, and is able to recover known sequence and structure motifs. Availability and implementation: pysster is freely available at https://github.com/budach/pysster. Supplementary information: Supplementary data are available at Bioinformatics online.

BACKGROUND: A large proportion of an organism's genome encodes for membrane proteins. Membrane proteins are important for many cellular processes, and several diseases can be linked to mutations in them. With the tremendous growth of sequence data, there is an increasing need to reliably identify membrane proteins from sequence, to functionally annotate them, and to correctly predict their topology. RESULTS: We introduce a technique called structural fragment clustering, which learns sequential motifs from 3D structural fragments. From over 500,000 fragments, we obtain 213 statistically significant, non-redundant, and novel motifs that are highly specific to alpha-helical transmembrane proteins. From these 213 motifs, 58 of them were assigned to function and checked in the scientific literature for a biological assessment. Seventy percent of the motifs are found in co-factor, ligand, and ion binding sites, 30% at protein interaction interfaces, and 12% bind specific lipids such as glycerol or cardiolipins. The vast majority of motifs (94%) appear across evolutionarily unrelated families, highlighting the modularity of functional design in membrane proteins. We describe three novel motifs in detail: (1) a dimer interface motif found in voltage-gated chloride channels, (2) a proton transfer motif found in heme-copper oxidases, and (3) a convergently evolved interface helix motif found in an aspartate symporter, a serine protease, and cytochrome b. CONCLUSIONS: Our findings suggest that functional modules exist in membrane proteins, and that they occur in completely different evolutionary contexts and cover different binding sites. Structural fragment clustering allows us to link sequence motifs to function through clusters of structural fragments. The sequence motifs can be applied to identify and characterize membrane proteins in novel genomes.

MOTIVATION: Misfolding of membrane proteins plays an important role in many human diseases such as retinitis pigmentosa, hereditary deafness and diabetes insipidus. Little is known about membrane proteins as there are only very few high-resolution structures. Single-molecule force spectroscopy is a novel technique, which measures the force necessary to pull a protein out of a membrane. Such force curves contain valuable information on the protein structure, conformation, and inter- and intra-molecular forces. High-throughput force spectroscopy experiments generate hundreds of force curves including spurious ones and good curves, which correspond to different unfolding pathways. Manual analysis of these data is a bottleneck and source of inconsistent and subjective annotation. RESULTS: We propose a novel algorithm for the identification of spurious curves and curves representing different unfolding pathways. Our algorithm proceeds in three stages: first, we reduce noise in the curves by applying dimension reduction; second, we align the curves with dynamic programming and compute pairwise distances and third, we cluster the curves based on these distances. We apply our method to a hand-curated dataset of 135 force curves of bacteriorhodopsin mutant P50A. Our algorithm achieves a success rate of 81% distinguishing spurious from good curves and a success rate of 76% classifying unfolding pathways. As a result, we discuss five different unfolding pathways of bacteriorhodopsin including three main unfolding events and several minor ones. Finally, we link folding barriers to the degree of conservation of residues. Overall, the algorithm tackles the force spectroscopy bottleneck and leads to more consistent and reproducible results paving the way for high-throughput analysis of structural features of membrane proteins.