seqgra: principled selection of neural network architectures for genomics prediction tasks
Abstract Motivation: Sequence models based on deep neural networks have achieved state-of-the-art performance on regulatory genomics prediction tasks, such as chromatin accessibility and transcription factor binding. But despite their high accuracy, their contributions to a mechanistic understandi...
| Main Authors: | , , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
Oxford University Press (OUP)
2022
|
| Online Access: | https://hdl.handle.net/1721.1/143575 |
| _version_ | 1826217204465532928 |
|---|---|
| author | Krismer, Konstantin Hammelman, Jennifer Gifford, David K |
| author2 | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory |
| author_facet | Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Krismer, Konstantin Hammelman, Jennifer Gifford, David K |
| author_sort | Krismer, Konstantin |
| collection | MIT |
| description | Abstract
Motivation: Sequence models based on deep neural networks have achieved state-of-the-art performance on regulatory genomics prediction tasks, such as chromatin accessibility and transcription factor binding. But despite their high
accuracy, their contributions to a mechanistic understanding of the biology of regulatory elements is often hindered
by the complexity of the predictive model and thus poor interpretability of its decision boundaries. To address this, we
introduce seqgra, a deep learning pipeline that incorporates the rule-based simulation of biological sequence data and
the training and evaluation of models, whose decision boundaries mirror the rules from the simulation process.
Results: We show that seqgra can be used to (i) generate data under the assumption of a hypothesized model of
genome regulation, (ii) identify neural network architectures capable of recovering the rules of said model and (iii)
analyze a model’s predictive performance as a function of training set size and the complexity of the rules behind
the simulated data.
Availability and implementation: The source code of the seqgra package is hosted on GitHub (https://github.com/gif
ford-lab/seqgra). seqgra is a pip-installable Python package. Extensive documentation can be found at https://
kkrismer.github.io/seqgra. |
| first_indexed | 2024-09-23T16:59:41Z |
| format | Article |
| id | mit-1721.1/143575 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T16:59:41Z |
| publishDate | 2022 |
| publisher | Oxford University Press (OUP) |
| record_format | dspace |
| spelling | mit-1721.1/1435752023-02-13T21:03:45Z seqgra: principled selection of neural network architectures for genomics prediction tasks Krismer, Konstantin Hammelman, Jennifer Gifford, David K Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Massachusetts Institute of Technology. Department of Biological Engineering Massachusetts Institute of Technology. Computational and Systems Biology Program Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science Abstract Motivation: Sequence models based on deep neural networks have achieved state-of-the-art performance on regulatory genomics prediction tasks, such as chromatin accessibility and transcription factor binding. But despite their high accuracy, their contributions to a mechanistic understanding of the biology of regulatory elements is often hindered by the complexity of the predictive model and thus poor interpretability of its decision boundaries. To address this, we introduce seqgra, a deep learning pipeline that incorporates the rule-based simulation of biological sequence data and the training and evaluation of models, whose decision boundaries mirror the rules from the simulation process. Results: We show that seqgra can be used to (i) generate data under the assumption of a hypothesized model of genome regulation, (ii) identify neural network architectures capable of recovering the rules of said model and (iii) analyze a model’s predictive performance as a function of training set size and the complexity of the rules behind the simulated data. Availability and implementation: The source code of the seqgra package is hosted on GitHub (https://github.com/gif ford-lab/seqgra). seqgra is a pip-installable Python package. Extensive documentation can be found at https:// kkrismer.github.io/seqgra. 2022-06-28T16:33:56Z 2022-06-28T16:33:56Z 2022-04-28 2022-06-28T13:58:21Z Article http://purl.org/eprint/type/JournalArticle https://hdl.handle.net/1721.1/143575 Krismer, Konstantin, Hammelman, Jennifer and Gifford, David K. 2022. "seqgra: principled selection of neural network architectures for genomics prediction tasks." Bioinformatics, 38 (9). en 10.1093/bioinformatics/btac101 Bioinformatics Creative Commons Attribution 4.0 International license https://creativecommons.org/licenses/by/4.0/ application/pdf Oxford University Press (OUP) Oxford University Press |
| spellingShingle | Krismer, Konstantin Hammelman, Jennifer Gifford, David K seqgra: principled selection of neural network architectures for genomics prediction tasks |
| title | seqgra: principled selection of neural network architectures for genomics prediction tasks |
| title_full | seqgra: principled selection of neural network architectures for genomics prediction tasks |
| title_fullStr | seqgra: principled selection of neural network architectures for genomics prediction tasks |
| title_full_unstemmed | seqgra: principled selection of neural network architectures for genomics prediction tasks |
| title_short | seqgra: principled selection of neural network architectures for genomics prediction tasks |
| title_sort | seqgra principled selection of neural network architectures for genomics prediction tasks |
| url | https://hdl.handle.net/1721.1/143575 |
| work_keys_str_mv | AT krismerkonstantin seqgraprincipledselectionofneuralnetworkarchitecturesforgenomicspredictiontasks AT hammelmanjennifer seqgraprincipledselectionofneuralnetworkarchitecturesforgenomicspredictiontasks AT gifforddavidk seqgraprincipledselectionofneuralnetworkarchitecturesforgenomicspredictiontasks |