Coordinate- and Sequence-Based Features for a new Combined Annotation-Dependent Depletion Framework of Structural Variants (CADD-SV v2.0)

Long-read based detection of large copy number variants with potential functional significance using the ContextSV structural variant caller

Abstract Long-read sequencing enables improved detection of structural variants (SVs) in the human genome due to its substantially increased read lengths. However, currently widely used long-read SV callers primarily rely on alignment-based evidence, limiting their ability to detect large and complex SVs and potentially missing disease-relevant events. To address these limitations, we developed ContextSV, a framework that integrates alignment evidence with copy number predictions derived from sequencing coverage and single-nucleotide variant allele frequencies to improve SV detection, particularly for large copy number variants (CNVs). We additionally developed ContextScore, a machine learning–based classification model to assign SV confidence scores based on genomic context features and integrated it within ContextSV. Through benchmarking analyses on both simulated and real datasets, we demonstrate that ContextSV improves detection of large CNVs and inversions that may be missed by existing long-read SV callers. We further illustrate its utility by identifying and experimentally validating multiple large SVs in the KOLF2.1J reference stem cell line that were not detected by other methods. Collectively, our results demonstrate that ContextSV serves as a valuable complement to existing long-read SV detection approaches by improving sensitivity for large and clinically relevant SVs.

A novel benchmark dataset for enzyme function prediction reveals the limitations of state-of-the-art models

It is demonstrated that modern EC predictors largely fail to distinguish catalytically incompetent variants from functional enzymes, and it is proposed that integrating structure-aware negative examples into both training and benchmarking is critical for developing functionally robust models in computational enzymology.

A high-resolution human pangenome structural variant resource for improved disease association

This work describes the full spectrum of genetic variation and shows that while 99% of the variants between any two genomes are single base-pair substitutions, 88% of the euchromatic variant base pairs are SVs, including insertions, deletions, duplications, and inversions.

INDELVAR: structure-informed prediction of in-frame indel pathogenicity with calibrated PP3/BP4 thresholds

DNCLA: A Deep Learning Model for TFBS Identification Based on Structural and Conformational Properties of Nucleotides and Dinucleotides

Identifying transcription factor binding sites (TFBSs) is fundamental to understanding complex gene regulatory mechanisms and the functions of non-coding regions. Although existing methods have achieved substantial strides, capturing both local structural features and long-range spatial dependencies within DNA sequences remains a major challenge for improving prediction accuracy. In this study, we propose DNCLA, a deep learning model that synergizes multisize convolutional fusion, Bidirectional Long ShortTerm Memory (Bi-LSTM) networks, and a multi-head self-attention mechanism. At the feature extraction level, DNCLA breaks through the limitations of traditional single-sequence encoding by fusing Nucleotide Chemical Properties (NCP) with Dinucleotide Physicochemical Properties (DPCP). NCP provides a refined characterization of chemical differences between bases based on ring structures, hydrogen bond sites, and functional group properties, while DPCP introduces parameters such as local structural stability and geometric flexibility of the DNA. Subsequently, the model extracts spatial evolution from these high-dimensional features through a multi-size convolutional module; captures long-range spatial dependencies using Bi-LSTM layers; and employs a multi-head self-attention mechanism to achieve adaptive weight distribution of global features, thereby enhancing the perception of key regulatory motifs. Results from training and testing the proposed model on 165 ChIPseq datasets demonstrate that DNCLA possesses robust generalization capabilities and high predictive performance in TFBSs identification. This suggests that the incorporation of physicochemical features better elucidates the essence of interactions between transcription factors and DNA.

Accurate de novo transcription unit annotation from run-on and sequencing data

Functional element annotations are critical tools used to provide insight into the molecular processes governing cell development, differentiation, and disease. Run-on and sequencing assays measure the production of nascent RNAs and can provide an effective data source for discovering functional elements. However, the accurate inference of functional elements from run-on sequencing data remains an open problem because the signal is noisy and challenging to model. Here we investigated computational approaches that convert run-on and sequencing data into annotations representing transcription units, including genes and non-coding RNAs. We developed a convolutional neural network, called convolutional discovery of gene anatomy using PRO-seq (CGAP), trained to identify different anatomical features of a transcription unit, which were then stitched together into transcript annotations using a hidden Markov model (HMM). Comparison with existing methods showed a significant performance improvement using our novel CGAP-HMM approach. We developed a voting system that ensembles the top three annotation strategies, resulting in large and significant improvements in transcription unit annotation accuracy over the best performing individual method. Finally, we also report a conditional generative adversarial network (cGAN) as a generative approach to transcription unit annotation that shows promise for further development. Collectively our work provides novel tools for de novo transcription unit annotation from run-on and sequencing data that are accurate enough to be useful in many applications.