BDGP Neural Network Sequence Tools

This repository preserves the splice-site and promoter prediction services developed for the Berkeley Drosophila Genome Project. The predictors and neural network models were authored by Dr. Martin G. Reese, who generously allowed the tools to be shared for non-commercial research use. This repo bundles the original CGI applications, the trained binaries, and a modern Docker wrapper so the tools can still be hosted for the research community.

Quick start (web service)

cd fruitfly
./run.sh                   # builds the image if needed and launches docker compose
# optionally set the hostname used by Apache and the self-signed cert
./run.sh --server-name myhost.local

After startup, visit:

• https://<server-name>:8080/seq_tools/splice.html
• https://<server-name>:8080/seq_tools/promoter.html

Accept the self-signed certificate when prompted. To stop the stack run docker compose down. Apache access and error logs are available through docker logs fruitfly.

Repository layout

• fruitfly/ – Deployable web root (CGI scripts, helper binaries, HTML, Apache configuration, Docker files).
• C/ – C source for the fa2* splice-site predictors and the nnpp promoter predictor, together with the saved network weights.
• snns/ – Stuttgart Neural Network Simulator snapshot used to train/export the network files.

Rebuilding the neural networks

The binaries in fruitfly/cgi-bin/seq_tools/genotator/ were built with the code in C/ and network definitions generated by SNNS. To produce your own models:

1. Install SNNS from the snns/ directory (or a compatible release).
2. Train the desired networks and export updated myNetwork.* files for each predictor (donor/acceptor, Drosophila/non-Drosophila, prokaryote/eukaryote).
3. Recompile the predictors in the C/ tree so they pick up the new network weights.
4. Replace the executables under fruitfly/cgi-bin/seq_tools/genotator/ and rebuild the Docker image.

Please retain the original copyright notices when redistributing modified code.