Dataset Information

GkmExplain: fast and accurate interpretation of nonlinear gapped k-mer SVMs.

ABSTRACT: SUMMARY:Support Vector Machines with gapped k-mer kernels (gkm-SVMs) have been used to learn predictive models of regulatory DNA sequence. However, interpreting predictive sequence patterns learned by gkm-SVMs can be challenging. Existing interpretation methods such as deltaSVM, in-silico mutagenesis (ISM) or SHAP either do not scale well or make limiting assumptions about the model that can produce misleading results when the gkm kernel is combined with nonlinear kernels. Here, we propose GkmExplain: a computationally efficient feature attribution method for interpreting predictive sequence patterns from gkm-SVM models that has theoretical connections to the method of Integrated Gradients. Using simulated regulatory DNA sequences, we show that GkmExplain identifies predictive patterns with high accuracy while avoiding pitfalls of deltaSVM and ISM and being orders of magnitude more computationally efficient than SHAP. By applying GkmExplain and a recently developed motif discovery method called TF-MoDISco to gkm-SVM models trained on in vivo transcription factor (TF) binding data, we recover consolidated, non-redundant TF motifs. Mutation impact scores derived using GkmExplain consistently outperform deltaSVM and ISM at identifying regulatory genetic variants from gkm-SVM models of chromatin accessibility in lymphoblastoid cell-lines. AVAILABILITY AND IMPLEMENTATION:Code and example notebooks to reproduce results are at https://github.com/kundajelab/gkmexplain. SUPPLEMENTARY INFORMATION:Supplementary data are available at Bioinformatics online.

SUBMITTER: Shrikumar A

PROVIDER: S-EPMC6612808 | biostudies-other | 2019 Jul

REPOSITORIES: biostudies-other

ACCESS DATA

Publications

GkmExplain: fast and accurate interpretation of nonlinear gapped k-mer SVMs.

Shrikumar Avanti A Prakash Eva E Kundaje Anshul A

Bioinformatics (Oxford, England) 20190701 14

<h4>Summary</h4>Support Vector Machines with gapped k-mer kernels (gkm-SVMs) have been used to learn predictive models of regulatory DNA sequence. However, interpreting predictive sequence patterns learned by gkm-SVMs can be challenging. Existing interpretation methods such as deltaSVM, in-silico mutagenesis (ISM) or SHAP either do not scale well or make limiting assumptions about the model that can produce misleading results when the gkm kernel is combined with nonlinear kernels. Here, we propo ...[more]

PMID: 31510661

Similar Datasets

Project description:A methodological survey of density functional theory (DFT) methods for the prediction of UV-visible (vis)-near-infrared (NIR) spectra of phthalocyanines is reported. Four methods, namely, full time-dependent (TD)-DFT and its Tamm-Dancoff approximation (TDA), together with their simplified modifications (sTD-DFT and sTDA, respectively), were tested by using the examples of unsubstituted and alkoxy-substituted metal-free ligands and zinc complexes. The theoretical results were compared with experimental data derived from UV-visible absorption and magnetic circular dichroism spectroscopy. Seven popular exchange-correlation functionals (BP86, B3LYP, TPSSh, M06, CAM-B3LYP, LC-BLYP, and ωB97X) were tested within these four approaches starting at a relatively modest level using 6-31G(d) basis sets and gas-phase BP86/def2-SVP optimized geometries. A gradual augmentation of the computational levels was used to identify the influence of starting geometry, solvation effects, and basis sets on the results of TD-DFT and sTD-DFT calculations. It was found that although these factors do influence the predicted energies of the vertical excitations, they do not affect the trends predicted in the spectral properties across series of structurally related substituted free bases and metallophthalocyanines. The best accuracy for the gas-phase vertical excitations was observed in the lower-energy Q-band region for calculations that made use of range-separated hybrids for both full and simplified TD-DFT approaches. The CAM-B3LYP functional provided particularly accurate results in the context of the sTD-DFT approach. The description of the higher-energy B-band region is considerably less accurate, and this demonstrates the need for further advances in the accuracy of theoretical calculations. Together with a general increase in accuracy, the application of simplified TD-DFT methods affords a 2-3 orders of magnitude speedup of the calculations in comparison to the full TD-DFT approach. It is anticipated that this approach will be widely used on desktop computers during the interpretation of UV-vis-NIR spectra of phthalocyanines and related macrocycles in the years ahead.

Dataset Information

GkmExplain: fast and accurate interpretation of nonlinear gapped k-mer SVMs.

Publications

GkmExplain: fast and accurate interpretation of nonlinear gapped k-mer SVMs.

Similar Datasets

OmicsDI is part of the ELIXIR infrastructure

Tweets