bioRxiv Science⌕ Search
Home/Archive/bioRxiv

bioRxiv · 10.1101/2021.05.23.445347

DLPacker: Deep Learning for Prediction of Amino Acid Side Chain Conformations in Proteins

Misiura, M.·Shroff, R.·Thyer, R.·Kolomeisky, A.
View original source ↗Export metadata ↓

Abstract

AO_SCPLOWBSTRACTC_SCPLOWPrediction of side chain conformations of amino acids in proteins (also termed packing) is an important and challenging part of protein structure prediction with many interesting applications in protein design. A variety of methods for packing have been developed but more accurate ones are still needed. Machine learning (ML) methods have recently become a powerful tool for solving various problems in diverse areas of science, including structural biology. In this work we evaluate the potential of Deep Neural Networks (DNNs) for prediction of amino acid side chain conformations. We formulate the problem as image-to-image transformation and train a U-net style DNN to solve the problem. We show that our method outperforms other physics-based methods by a significant margin: reconstruction RMSDs for most amino acids are about 20% smaller compared to SCWRL4 and Rosetta Packer with RMSDs for bulky hydrophobic amino acids Phe, Tyr and Trp being up to 50% smaller.

Source connections

Explore related subjects

biophysicsPreprint

Keep this discovery

Explore connections, maps & timelines

BibTeXRIS

Misiura, M., Shroff, R., Thyer, R., Kolomeisky, A.. 2021-05-25. DLPacker: Deep Learning for Prediction of Amino Acid Side Chain Conformations in Proteins. https://doi.org/10.1101/2021.05.23.445347

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

KEEP EXPLORING

Related preprints

bioRxiv2026

Expression-System Biases Compromise the Functional Assessment of KCNH2 Variants and Clinical Risk Assessment

Background: Functional characterization of KCNH2 variants is essential for assessing pathogenicity in long QT syndrome type 2. Because hERG channels assemble as tetramers, the relative expression of wild-type (WT) and variant subunits critically influences channel function and pathogenicity assessment. However, commonly used experimental systems may not reproduce the balanced allelic expression observed in heterozygous patients. Methods: Allele-specific KCNH2 expression was analyzed in two independent RNA-sequencing datasets from human left ventricular tissue. Using the dominant-negative T74R hERG variant as a model, we calibrated the relationship between WT-to-variant expression ratios and hERG current amplitude. Using automated patch-clamp electrophysiology, we compared 3 dual-expression strategies: internal ribosome entry site (IRES), P2A-linked tandem constructs, and dual cytomegalovirus promoter (dCMV) constructs. Co-electroporation of WT and variant plasmids in different ratios was the reference. Results: Allele-specific expression analysis supported balanced KCNH2 expression in human left ventricular tissue. At a 1:1 WT-to-T74R expression ratio, hERG current amplitude was reduced to 31.0% of WT, consistent with a dominant-negative mechanism in which incorporation of 2 or more T74R subunits renders the channel nonfunctional. All 3 dual-expression strategies produced substantial expression imbalances, with the first cassette consistently favored. IRES constructs showed the greatest asymmetry. P2A constructs partially improved expression balance but introduced an additional N-terminal proline in the second cassette altering T74R-associated trafficking and channel biophysics. The dCMV system also failed to achieve balanced expression, including after stable integration. Co-electroporation of separate WT and variant plasmids provided the most reliable means of controlling their relative expression. Conclusions: Commonly used dual-expression constructs do not reliably reproduce balanced WT and variant hERG expression, potentially distorting assessments of dominant-negative effects and variant pathogenicity. Co-electroporation of separate plasmids provides a robust approach for high-throughput functional phenotyping under heterozygous conditions, with implications for the interpretation of KCNH2 variants and potentially other genes encoding multimeric ion channels.

biophysics↗
bioRxiv2026

Learning from Hi-C reveals nuclear dynamical reorganization in budding yeast across the cell cycle

Hi-C contact maps provide population-averaged measurements of chromosomal contacts. Converting these measurements, however, into three-dimensional structures and associated dynamics remains an underdetermined problem. Existing approaches broadly fall into two classes. Data-driven reconstruction methods that infer either a consensus structure or an ensemble of conformations, but do not inform on chromosome dynamics. Physics-based models, which generate dynamical trajectories, but require prior assumptions about nuclear geometry, chromatin mechanics, tethering, or molecular interactions. Here, we bridge these approaches using the Full-Inversion Chromatin model, a data-driven whole-genome polymer framework that learns an effective force field directly from Hi-C contact probabilities, integrated with Serpentine, a tool to quantify contact-map discrepancies while accounting for differences in sequencing coverage and sparsity. This generates structural ensembles of chromosomes and effective chromatin dynamics. We apply this method at 5-kb resolution to Saccharomyces cerevisiae in G1, mid-S, S/G2, G2/M, anaphase, and quiescence. Without explicitly imposing the spindle pole body, centromere or telomere tethering, predefined chromatin states, or nuclear-envelope interactions, the inferred ensembles recover these major features of yeast nuclear organization. Importantly, although dynamical information is not used during model training, the resulting polymer trajectories reproduce the subdiffusive scaling and chromosomal variations of independently measured chromatin motion. The model further predicts heterogeneous mobility along chromosomes and a distinctive telomere-dependent dynamical organization. Thus, population-averaged Hi-C contacts contain sufficient information to infer not only three-dimensional genome organization, but also effective, locus-dependent chromatin dynamics.

biophysics↗
bioRxiv2026

One high-order ion-coupling mechanism generates multiple SLC1A stoichiometries

Human excitatory amino acid transporter 2 (hEAAT2) is a glial transporter that clears glutamate and aspartate from the extracellular space to terminate neurotransmission. When this essential role is disrupted, excess extracellular glutamate causes excitotoxicity and neurological disease. hEAAT2 transports amino acids into cells against their concentration gradients by coupling uptake to the co-transport of Na+ and H+, and counter-transport of K+. Substrate transport also activates a thermodynamically uncoupled Cl- conductance, whose significance is unknown. Here, to dissect the molecular mechanisms underlying coupled and uncoupled ion transport by EAATs, we leverage the tractable Drosophila melanogaster homologue dEAAT2, which transports taurine as well as glutamate and aspartate. By functionally characterising dEAAT2, we confirmed it has features of a canonical EAAT transport cycle, including an uncoupled Cl- conductance. However, it functions an ion-independent electroneutral exchanger, in stark contrast to its ion-coupled electrogenic human counterpart. Cryo-electron microscopy (cryo-EM) and computational simulations confirmed the inability of dEAAT2 to transport cations, but just two point mutations conferred Na+, H+, and K+ dependence, reconstituting the complex ion coupling profile of human transporters in the Drosophila exchanger. We also showed that dEAAT2 transports taurine in a complex with Cl-, bestowing a unique Cl--dependence on taurine transport, independent of uncoupled Cl- flux. Our findings show that ancestral exchangers in the EAAT family preserve the machinery to achieve complex ion-coupled transport, revealing the molecular basis of high-order ion coupling in hEAATs.

biophysics↗