Kipoi-Accelerating The Community Exchange and Reuse of Predictive Models for Genomics

• Abhimanyu Banerjee  |  Anshul Kundaje  |  Avanti Shrikumar  |  Daniel S. Kim  |  Johnny Israeli  |  Julien Gagneur  |  Jun Cheng  |  Lara Urban  |  Nancy Xu  |  Oliver Stegle  |  Roman Kreuzhuber  |  Žiga Avsec

• Technical University of Munich Germany

Advanced machine learning models applied to large-scale genomics datasets hold the promise to be major drivers for genome science. Once trained, such models can serve as a tool to probe the relationships between data modalities, including the effect of genetic variants on phenotype. ClinVar - variants in the range [-40nt, 10nt] around the splicing acceptor or variants in the range [-10, 10] nt around the splice donor of a protein coding gene were used to train. Only variants labelled 'Pathogenic' or 'Benign' were used. Data from all chromosomes was used for training. Logistic regression implemented in scikit-learn with default parameters was used to build the meta model using different feature subsets. 10-fold cross-validation was used to evaluate models using the auROC metric.

Input variables : Combined Annotation-Dependent Depletion score, CADD phred-like rank score based on whole genome CADD raw scores, phyloP (phylogenetic p-values) conservation score based on the multiple alignments of 33 placental mammal genomes including human, conservation score based on the multiple alignments of 10 primate genomes including human etc.
Output Variables : Clinical significance: Benign/pathogenic

Visit Model : kipoi.org

Additional links : biorxiv.org | github.com