Published December 5, 2023
| Version v2
Dataset
Open
SQUID manuscript workflow with outputs
Description
This repository reproduces the analysis performed in for our method SQUID (Surrogate Quantitative Interpretability for Deepnets; Seitz, McCandlish, Kinney and Koo). It contains tools to apply SQUID on several previously-published genomic models, and compare its results to existing attribution methods.
The scripts contained in this release are a replica of those found in the current release of our GitHub repository (https://github.com/evanseitz/squid-manuscript as of commit 5b47a7d), with the addition that all intermediate and final outputs are provided here.
Files
deepstarr.model.json
Files
(26.1 GB)
| Name | Size | Download all |
|---|---|---|
|
md5:0d735698d9419dbabfd603f7e6ed2d2c
|
162.5 MB | Download |
|
md5:7e53a9351b2520a4713a5ffdb5f1566c
|
2.6 MB | Download |
|
md5:9b796f79441e53dc75dd79b911fff872
|
11.6 kB | Preview Download |
|
md5:beb4fb9a50ad99531ad75fd60e323690
|
128.3 MB | Download |
|
md5:45c10e8f65ec7ac002caa9ff03aea5db
|
3.5 GB | Preview Download |
|
md5:34ff0b87cf93a9de654be6451602392a
|
502.9 MB | Preview Download |
|
md5:00560fa5312ed33081b332bf869cc60b
|
5.1 GB | Preview Download |
|
md5:87c0f59085a50a24ab644c28025b1143
|
1.9 GB | Preview Download |
|
md5:9c8e21938b99107e17cecf55396dbae0
|
12.2 GB | Preview Download |
|
md5:e06f5910726fe90069bbb001f770d848
|
907.0 MB | Preview Download |
|
md5:de455622051f965ccbc7e3242dfd7633
|
859 Bytes | Download |
|
md5:21cc8a4552210aa502ae350bc995d8f9
|
77.3 MB | Download |
|
md5:891c035ddefaebc9bf31ea6659027124
|
808.6 MB | Preview Download |
|
md5:891c035ddefaebc9bf31ea6659027124
|
808.6 MB | Preview Download |
Additional details
Related works
- Is variant form of
- Workflow: https://github.com/evanseitz/squid-manuscript (URL)
Dates
- Updated
-
2023-12-05Added front-facing SQUID repository (i.e., "squid-nn.zip"). Within the front-facing repository, included outputs for BPNet-based global surrogate modeling shown in paper, and removed older version (i.e., "examples_BPNet_global.zip"). Revised "squid-manuscript.zip" to mirror outputs presented in bioRxiv release; exact changes can be seen in the following commits: (1) https://github.com/evanseitz/squid-manuscript/commit/66367d087acb9716211497caa178be3b3d2f4846; (2): https://github.com/evanseitz/squid-manuscript/commit/5b47a7d23341fd9e7db6ab9ed321945e3fd8d18f. Some genomic sequence datasets and DNNs have also been copied into the main directory, for ease of downloading in external Colab notebooks.