Deep Learning Implicitly Handles Tissue Specific Phenomena to Predict Tumor DNA Accessibility and Immune Activity

Wnuk, Kamil; Sudol, Jeremi; Givechian, Kevin B.; Soon‐Shiong, Patrick; Rabizadeh, Shahrooz; Szeto, Christopher; Vaske, Charles J.

doi:10.1016/j.isci.2019.09.018

Cited by 7 publications

(4 citation statements)

References 38 publications

(49 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…While it may be tempting to conclude from this that it is safe to use GradCAM if we confine ourselves to architectures that have longer filters in the first layer, doing so would come at the expense of model performance; trends in both computer vision [28] and in regulatory genomics [29] have found that performance can be improved by “factorizing” layers that have large convolutional kernels into several layers that have smaller convolutional kernels. While it is tempting to think that we can safely trade of model performance in exchange for improved interpretability, the reality is that the two are not completely independent; for example, in our study, we found that the best explanations across all methods were derived from the better-performing DeepSEA-initialized models ( Fig.…”

Section: Resultsmentioning

confidence: 99%

Towards More Realistic Simulated Datasets for Benchmarking Deep Learning Models in Regulatory Genomics

Prakash

Shrikumar

Kundaje

2021

Preprint

View full text Add to dashboard Cite

Deep neural networks and support vector machines have been shown to accurately predict genomewide signals of regulatory activity from raw DNA sequences. These models are appealing in part because they can learn predictive DNA sequence features without prior assumptions. Several methods such as in-silico mutagenesis, GradCAM, DeepLIFT, Integrated Gradients and GkmExplain have been developed to reveal these learned features. However, the behavior of these methods on regulatory genomic data remains an area of active research. Although prior work has benchmarked these methods on simulated datasets with known ground-truth motifs, these simulations employed highly simplified regulatory logic that is not representative of the genome. In this work, we propose a novel pipeline for designing simulated data that comes closer to modeling the complexity of regulatory genomic DNA. We apply the pipeline to build simulated datasets based on publicly-available chromatin accessibility experiments and use these datasets to benchmark different interpretation methods based on their ability to identify ground-truth motifs. We find that a GradCAM-based method, which was reported to perform well on a more simplified dataset, does not do well on this dataset (particularly when using an architecture with shorter convolutional kernels in the first layer), and we theoretically show that this is expected based on the nature of regulatory genomic data. We also show that Integrated Gradients sometimes performs worse than gradient-times-input, likely owing to its linear interpolation path. We additionally explore the impact of user-defined settings on the interpretation methods, such as the choice of "reference"/"baseline", and identify recommended settings for genomics. Our analysis suggests several promising directions for future research on these model interpretation methods. Code and links to data are available at https://github.com/kundajelab/interpret-benchmark.

show abstract

Section: Resultsmentioning

confidence: 99%

Towards More Realistic Simulated Datasets for Benchmarking Deep Learning Models in Regulatory Genomics

Prakash

Shrikumar

Kundaje

2021

Preprint

View full text Add to dashboard Cite

show abstract

“…After the model is trained using data from multiple cell types, it can then be used to predict open chromatin in new cell types. This approach has demonstrated good prediction performance in both Wnuk et al (2019) and Nair et al (2019).…”

Section: Predicting Chromatin Accessibility Using Gene Expression and Dna Sequencesmentioning

confidence: 95%

“…Recently, Wnuk et al (2019) and Nair, Kim, Perricone, and Kundaje (2019) (ChromDragoNN) explored the idea of combining gene expression with DNA sequence to predict chromatin accessibility. Both studies used an approach similar to Basset to process the DNA sequence information based on CNN.…”

Section: Predicting Chromatin Accessibility Using Gene Expression and Dna Sequencesmentioning

confidence: 99%

Genome‐wide prediction of chromatin accessibility based on gene expression

Zhou

2020

WIREs Computational Stats

View full text Add to dashboard Cite

Decoding gene regulation in a biological system requires information from both transcriptome and regulome. While multiple high‐throughput transcriptome and regulome mapping technologies are available, transcriptome profiling is more widely used. Today, over a million bulk and single‐cell gene expression samples are publicly available. This number is orders of magnitude larger than the number of available regulome samples. Most of the gene expression samples do not have corresponding regulome data. However, it is possible to obtain regulome information via prediction. Open chromatin is a hallmark of active regulatory elements. This mini‐review discusses recent advances in predicting chromatin accessibility using gene expression data, including both the development of prediction methods and their applications in expanding the regulome catalog, improving regulome analysis, integrating transcriptome and regulome data, and facilitating single‐cell analysis of gene regulation. This article is categorized under: Applications of Computational Statistics > Genomics/Proteomics/Genetics Data: Types and Structure > Massive Data Statistical Models > Linear Models

show abstract

“…We benchmarked fastISM against a standard implementation of ISM. We choose 3 types of models that take DNA sequence as input-the Basset architecture (Kelley et al, 2016), the Factorized Basset architecture (Wnuk et al, 2019) and the BPNet architecture (Avsec et al, 2019). The first two models output scalar values for each output task, whereas the BPNet model outputs a profile vector of length equal to the input sequence length, and a scalar count.…”

Section: In-silico Saturation Mutagenesis Formentioning

confidence: 99%

fastISM: performantin silicosaturation mutagenesis for convolutional neural networks

et al. 2022

View full text Add to dashboard Cite

Motivation Deep learning models such as convolutional neural networks are able to accurately map biological sequences to associated functional readouts and properties by learning predictive de novo representations. In-silico saturation mutagenesis (ISM) is a popular feature attribution technique for inferring contributions of all characters in an input sequence to the model’s predicted output. The main drawback of ISM is its runtime, as it involves multiple forward propagations of all possible mutations of each character in the input sequence through the trained model to predict the effects on the output. Results We present fastISM, an algorithm that speeds up ISM by a factor of over 10x for commonly used convolutional neural network architectures. fastISM is based on the observations that the majority of computation in ISM is spent in convolutional layers, and a single mutation only disrupts a limited region of intermediate layers, rendering most computation redundant. fastISM reduces the gap between backpropagation-based feature attribution methods and ISM. It far surpasses the runtime of backpropagation-based methods on multi-output architectures, making it feasible to run ISM on a large number of sequences. Availability An easy-to-use Keras/TensorFlow 2 implementation of fastISM is available at https://github.com/kundajelab/fastISM. fastISM can be installed using pip install fastism. A hands-on tutorial can be found at https://colab.research.google.com/github/kundajelab/fastISM/blob/master/notebooks/colab/DeepSEA.ipynb. Supplementary information Supplementary data are available at Bioinformatics online.

show abstract

Deep Learning Implicitly Handles Tissue Specific Phenomena to Predict Tumor DNA Accessibility and Immune Activity

Cited by 7 publications

References 38 publications

Towards More Realistic Simulated Datasets for Benchmarking Deep Learning Models in Regulatory Genomics

Towards More Realistic Simulated Datasets for Benchmarking Deep Learning Models in Regulatory Genomics

Genome‐wide prediction of chromatin accessibility based on gene expression

fastISM: performantin silicosaturation mutagenesis for convolutional neural networks

Contact Info

Product

Resources

About