Rank Projection Trees for Multilevel Neural Network Interpretation

A variety of methods have been proposed for interpreting nodes in deep neural networks, which typically involve scoring nodes at lower layers with respect to their effects on the output of higher-layer nodes (where lower and higher layers are closer to the input and output layers, respectively). However, we may be interested in picking out a prioritized collection of subsets of the inputs across a range of scales according to their importance for an output node, and not simply a prioritized ranking across the inputs as singletons. Such a situation may arise in biological applications, for instance, where we are interested in epistatic effects between groups of genes in determining a trait of interest. Here, we outline a flexible framework which may be used to generate multiscale network interpretations, using any previously defined scoring function. We demonstrate the ability of our method to pick out biologically important genes and gene sets in the domains of cancer and psychiatric genomics.

[1]  Ankur Taly,et al.  Gradients of Counterfactuals , 2016, ArXiv.

[2]  Andrew Zisserman,et al.  Deep Inside Convolutional Networks: Visualising Image Classification Models and Saliency Maps , 2013, ICLR.

[3]  Max Welling,et al.  Visualizing Deep Neural Network Decisions: Prediction Difference Analysis , 2017, ICLR.

[4]  Mingming Jia,et al.  COSMIC: somatic cancer genetics at high-resolution , 2016, Nucleic Acids Res..

[5]  Daniel H. Geschwind,et al.  Systems biology and gene networks in neurodevelopmental and neurodegenerative disorders , 2015, Nature Reviews Genetics.

[6]  Tom H. Pringle,et al.  The human genome browser at UCSC. , 2002, Genome research.

[7]  Michael J. Purcaro,et al.  The PsychENCODE project , 2015, Nature Neuroscience.

[8]  Nitesh V. Chawla,et al.  SMOTE: Synthetic Minority Over-sampling Technique , 2002, J. Artif. Intell. Res..

[9]  Anna Shcherbina,et al.  Not Just a Black Box: Learning Important Features Through Propagating Activation Differences , 2016, ArXiv.

[10]  Steve Horvath,et al.  WGCNA: an R package for weighted correlation network analysis , 2008, BMC Bioinformatics.

[11]  David D. Cox,et al.  Hyperopt: A Python Library for Optimizing the Hyperparameters of Machine Learning Algorithms , 2013, SciPy.

[12]  The Icgctcga Pan-Cancer Analysis of Whole Genomes Consortium.  Pan-cancer analysis of whole genomes , 2020 .

[13]  Avanti Shrikumar,et al.  Learning Important Features Through Propagating Activation Differences , 2017, ICML.

[14]  Guangchuang Yu,et al.  clusterProfiler: an R package for comparing biological themes among gene clusters. , 2012, Omics : a journal of integrative biology.

[15]  S. Horvath,et al.  A General Framework for Weighted Gene Co-Expression Network Analysis , 2005, Statistical applications in genetics and molecular biology.

[16]  Icgc,et al.  Pan-cancer analysis of whole genomes , 2017, bioRxiv.

[17]  O. Troyanskaya,et al.  Predicting effects of noncoding variants with deep learning–based sequence model , 2015, Nature Methods.

[18]  Bronwen L. Aken,et al.  GENCODE: The reference human genome annotation for The ENCODE Project , 2012, Genome research.