DNCON2: Improved protein contact prediction using two-level deep convolutional neural networks

Adhikari, B.; Hou, J.; Cheng, J.

doi:10.1101/222893

Home/Archive/bioRxiv

bioRxiv · 10.1101/222893

DNCON2: Improved protein contact prediction using two-level deep convolutional neural networks

Adhikari, B.·Hou, J.·Cheng, J.

View original source ↗Export metadata ↓

Abstract

MotivationSignificant improvements in the prediction of protein residue-residue contacts are observed in the recent years. These contacts, predicted using a variety of coevolution-based and machine learning methods, are the key contributors to the recent progress in ab initio protein structure prediction, as demonstrated in the recent CASP experiments. Continuing the development of new methods to reliably predict contact maps is essential to further improve ab initio structure prediction.\n\nResultsIn this paper we discuss DNCON2, an improved protein contact map predictor based on two-level deep convolutional neural networks. It consists of six convolutional neural networks - the first five predict contacts at 6, 7.5, 8, 8.5, and 10 [A] distance thresholds, and the last one uses these five predictions as additional features to predict final contact maps. On the free-modeling datasets in CASP10, 11, and 12 experiments, DNCON2 achieves mean precisions of 35%, 50%, and 53.4%, respectively, higher than 30.6% by MetaPSICOV on CASP10 dataset, 34% by MetaPSICOV on CASP11 dataset, and 46.3% by Raptor-X on CASP12 dataset, when top L/5 long-range contacts are evaluated. We attribute the improved performance of DNCON2 to the inclusion of short- and medium-range contacts into training, two-level approach to prediction, use of the state-of-the-art optimization and activation functions, and a novel deep learning architecture that allows each filter in a convolutional layer to access all the input features of a protein of arbitrary length.\n\nAvailabilityThe web server of DNCON2 is at http://sysbio.rnet.missouri.edu/dncon2/ where training and testing datasets as well as the predictions for CASP10, 11, and 12 free-modeling datasets can also be downloaded. Its source code is available at https://github.com/multicom-toolbox/DNCON2/.\n\nContactchengji@missouri.edu\n\nSupplementary informationSupplementary data are available online.

Source connections

Published article ↗

Explore related subjects

bioinformatics Preprint

Keep this discovery

Explore connections, maps & timelines

BibTeX RIS

Adhikari, B., Hou, J., Cheng, J.. 2017-11-21. DNCON2: Improved protein contact prediction using two-level deep convolutional neural networks. https://doi.org/10.1101/222893

Cite the original work for its findings. Save a collection to share your selection of sources.

KEEP EXPLORING

Related preprints

bioRxiv2026

scRM: A single-cell foundation model trained on scRNA-seq reads

To our knowledge, scRM is the first single-cell foundation model trained on sequencing reads, demonstrating that raw data pretraining is not only feasible but competitive at a fraction of the computational cost. We anticipate that scaling this framework and exploring alternative architectures will unlock biological information inaccessible to count models and extend the dataset-specific pretraining it already makes possible. The source code is available at https://github.com/altos-labs/scRM.

bioinformatics↗

bioRxiv2026

Local fractal dimension as an indicator of functional reorganization in molecular interaction networks

Changes in protein abundance incompletely describe how molecular perturbations reshape cellular organization. Here we introduce DIFFRACT, a framework that integrates quantitative proteomics with protein-protein interaction networks to identify reorganization through changes in local fractal dimension. Molecular associations define effective distances between connected proteins, allowing the geometry of their surrounding network neighborhoods to be compared across conditions. Analyses across four organisms (E. coli, A. thaliana, C. elegans, and H. sapiens) and three interaction resources (STRING, BioGRID, and IntAct) establish heterogeneous local scaling, while controlled perturbations demonstrate sensitivity to regional changes and invariance to uniform distance rescaling. Applied to RAVER1-deficient A549 cells, DIFFRACT identifies components associated with established RNA-regulatory functions and candidate roles in tRNA biology, mitochondrial organization and ciliogenesis. Selected RNA-regulatory and ciliary themes are not significantly enriched by abundance-ranked gene set analysis, supporting complementary information from network geometry. Selected functional annotations also recur in HEK293T cells. Guided by the ciliary predictions, we show that RAVER1 depletion increases the proportion of ciliated A549 cells 2.57-fold and increases cilium length. DIFFRACT thus connects changes in molecular network geometry to experimentally testable biological responses, extending the interpretation of proteomic perturbations beyond differential abundance.

bioinformatics↗

bioRxiv2026

Peripheral Exercise Transcriptomes Do Not Converge on Glutamatergic or Neuroinflammatory Signatures: A Cross-Platform Meta-Analysis

Executive dysfunction following physical overtraining has been linked to glutamatergic dysregulation in the lateral prefrontal cortex (LPFC), but whether this central mechanism is reflected in peripheral gene expression is unknown. This study tested whether peripheral transcriptomic responses to exercise converge on the glutamatergic or neuroinflammatory pathways implicated in LPFC dysfunction. Three publicly available GEO datasets spanning skeletal muscle, whole blood, and circulating neutrophils were analyzed across four exercise groups using limma, with a discovery threshold of nominal p < 0.05 and absolute log2 fold change above 0.5. Differential expression analysis identified 7,311 unique differentially expressed genes (DEGs) across the four contrasts, but overlap between datasets was limited: 7 genes were shared between skeletal muscle and blood, 38 between skeletal muscle and the GSE43856 contrasts, and 90 between blood and the GSE43856 contrasts, with none significant across all three studies. Gene Ontology and KEGG enrichment of the DEG union identified ribonucleoprotein complex biogenesis, ribosome biogenesis, protein folding, and T cell activation as the dominant biological processes; no glutamatergic or neuroinflammatory terms reached significance. Cross-tissue concordance between skeletal muscle and blood was absent (r = 0). Effect sizes among the genes driving the top enriched pathways were dominated by ribosomal proteins, heat shock proteins, and translation initiation factors rather than glutamate receptors or transporters. These findings indicate that peripheral transcriptomic responses to exercise are tissue-specific and dominated by proteostasis, translational, and immune programs rather than glutamatergic signaling. Direct assessment of central nervous system tissue or fluid-based glutamatergic biomarkers may be required to link overtraining-induced fatigue to LPFC dysfunction.

bioinformatics↗