Improving representations of genomic sequence motifs in convolutional networks with exponential activations

Abstract

Deep convolutional neural networks (CNNs) trained on regulatory genomic sequences tend to learn distributed feature representations across many filters, making it challenging to decipher biologically meaningful representations, such as sequence motifs. Here we perform a comprehensive analysis on synthetic sequences to investigate the role that CNN activations have on model interpretability. We introduce a novel application of the exponential activation that when applied to first layer filters, consistently leads to interpretable and robust representations of motifs compared to other commonly used activations, both qualitatively and quantitatively. Strikingly, we demonstrate that CNNs with better test performance do not necessarily imply more interpretable representations with attribution methods. We find that CNNs with exponential activations significantly improve the efficacy of a CNN's ability to recover biologically meaningful representations with attribution methods. We demonstrate these results generalize to real DNA sequences across several in vivo datasets. Together, this work demonstrates how a small modification to existing CNNs, i.e. setting exponential activations in the first layer, can significantly improve the robustness and interpretabilty of learned representations directly in convolutional filters and indirectly with attribution methods. 3/18 4/18

Global importance analysis: An interpretability method to quantify importance of genomic features in deep neural networks

“…To gain insights into what DNN-based methods have learned, DLPRB visualizes filter representations while cDeepbind employs in silico mutagenesis. Filter representations are sensitive to network design choices [ 29 , 30 ]; ResidualBind is not designed with the intention of learning interpretable filters. Hence, we opted to employ in silico mutagenesis, which systematically probes the effect size that each possible single nucleotide mutation in a given sequence has on model predictions.…”

“…For RBPs, this has been accomplished by visualizing first convolutional layer filters and via attribution methods [ 13 , 18 , 23 , 24 ]. First layer filters have been shown to capture motif-like representations, but their efficacy depends highly on choice of model architecture [ 29 ], activation function [ 30 ], and training procedure [ 31 ]. First-order attribution methods, including in silico mutagenesis [ 13 , 32 ] and other gradient-based methods [ 19 , 33 – 36 ], are interpretability methods that identify the independent importance of single nucleotide variants in a given sequence toward model predictions—not the effect size of extended patterns such as sequence motifs.…”

Global Importance Analysis: An Interpretability Method to Quantify Importance of Genomic Features in Deep Neural Networks

“…To gain insights into what DNN-based methods have learned, DLPRB visualizes filter representations while cDeepbind employs in silico mutagenesis. Filter representations are sensitive to network design choices 29,30 ; ResidualBind is not designed with the intention of learning interpretable filters. Hence, we opted to employ in silico mutagenesis, which systematically probes the effect size that each possible single nucleotide mutation in a given sequence has on model predictions.…”

“…For RBPs, this has been accomplished by visualizing first convolutional layer filters and via attribution methods 13,18,23,24 . First layer filters have been shown to capture motif-like representations, but their efficacy depends highly on choice of model architecture 29 , activation function 30 , and training procedure 31 . First-order attribution methods, including in silico mutagenesis 13,32 and other gradient-based methods 19,33–36 , are interpretability methods that identify the independent importance of single nucleotide variants in a given sequence toward model predictions – not the effect size of extended patterns such as sequence motifs.…”

fastISM: Performant in-silico saturation mutagenesis for convolutional neural networks

“…In comparison to most backpropagation-based methods that often use heuristic rules and approximations, ISM faithfully represents the model's response to mutations at individual positions. This makes it the method of choice when evaluating the effect of genetic variants on the output (Zhou and Troyanskaya, 2015;Zhou et al, 2018;Wesolowska-Andersen et al, 2020), and it is also used as a benchmark reference when evaluating fidelity of other feature attribution methods (Koo and Ploenzke, 2020). Unlike ISM, backpropagation-based methods like DeepLIFT and Integrated Gradients rely on a predefined set of "neutral" input sequences that are used as explicit references to estimate attribution scores.…”