Merge

.gitignore

@@ -1,5 +1,6 @@
+Reservoir
 RayLogs
 __pycache__
 *pyc
 *egg-info
-.DS_Store
+.DS_Store

README.md

@@ -1,23 +1,39 @@
-# RECOVER: sequential model optimization platform for combination drug repurposing identifies novel synergistic compounds *in vitro*
+# Machine Learning Driven Candidate Compound Generation for Drug Repurposing
+Based on RECOVER: sequential model optimization platform for combination drug repurposing identifies novel synergistic compounds *in vitro*
 [![DOI](https://zenodo.org/badge/320327566.svg)](https://zenodo.org/badge/latestdoi/320327566)
 
-RECOVER is a platform that can guide wet lab experiments to quickly discover synergistic drug combinations active 
-against a cancer cell line, requiring substantially less screening than an exhaustive evaluation 
-([preprint](https://arxiv.org/abs/2202.04202)).
+This repository is an implementation of RECOVER, a platform that can guide wet lab experiments to quickly discover synergistic drug combinations,
+([preprint](https://arxiv.org/abs/2202.04202)), howerver instead of using an ensemble model to get Synergy predictions with uncertainty, we used multiple realization of a Bayesian Neural Network model. 
+Since the weights are drawn from a distribution, they differ for every run of a trained model and hence give different results. The goal was to get a more precise uncertainty and achieve i quicker since the model doesn't have to be trained multiple times. 
 
+<p float="left">
+  <img src="docs/images/ProjectInfographics.png" alt="Overview" width="300"/>
+  <img src="docs/images/ModelOverview.png" alt="Model Overview" width="400"/>
+</p>
 
-![Overview](docs/images/overview.png "Overview")
+## Repository overview
+There are 4 branches in this Git Repository.
+
+**Master**, where the original RECOVER is stored with Bayesian setup but original config files, so the initial pipeline can be easily recreated.   
+**Bayesian_after_merge**, with a Bayesian model used only in the layers after the bilinear merge.   
+**Bayesian_before_and_after_merge**, with a Bayesian model used in the layers before and after the bilinear merge.  
+**Weight_uncertainty**, a Bayesian Neural Network library implemented from scratch and two priors introduced in order to optimize the model. (https://doi.org/10.48550/arXiv.1505.05424)  
+
+Every branch will be further explaned in its own README.
 
 ## Environment setup
 
-**Requirements**: Anaconda (https://www.anaconda.com/) and Git LFS (https://git-lfs.github.com/). Please make sure 
-both are installed on the system prior to running installation.
+**Requirements and Installation**: 
+For all the requirements and installation steps dataset and the project check the orginal RECOVER repository (https://github.com/RECOVERcoalition/Recover.git). 
 
-**Installation**: enter the command `source install.sh` and follow the instructions. This will create a conda 
-environment named **recover** and install all the required packages including the 
-[reservoir](https://github.com/RECOVERcoalition/Reservoir) package that stores the primary data acquisition scripts.
+**Potential issues and fixes**
+- Use the requirements file from this Repository
+- Cloning dataset: Makesure to install Git LFS (Since there are some files over 100 MB git-lfs is needed for cloning.)
+- Numpy 1.24 does not support - Even though there was no mention about the numpy version in original RECOVER requirements there are some compatibility issues when using other libraries therefore we have to specifically downgrade numpy to make it working without any issues. Try, pip install numpy==1.22.4
+- Ray tune issue: Use the updated train.py within this repo to avoid ray tune path issues when saving checkpoints
+- Issues with the config files: Check for such unused modules which are imported and remove them to avoid module not fount errors. 
+- rdkit installation path in original Repository is outdated. Try, conda install -c conda-forge rdkit 
 
-In case you have any issue with the installation procedure of the *reservoir*, you can access and download all files directly from this [google drive](https://drive.google.com/drive/folders/1MYeDoAi0-qnhSJTvs68r861iMOdoqYki?usp=share_link).
 
 ## Running the pipeline
 
@@ -31,9 +47,3 @@ For example, to run the pipeline with configuration from
 the file `model_evaluation.py`, run `python train.py --config model_evaluation`.
 
 Log files will automatically be created to save the results of the experiments.
-
-## Note
-
-This Recover repository is based on research funded by (or in part by) the Bill & Melinda Gates Foundation. The 
-findings and conclusions contained within are those of the authors and do not necessarily reflect positions or policies 
-of the Bill & Melinda Gates Foundation.

cluster_job.sh

@@ -0,0 +1,76 @@
+#!/bin/bash -l
+# shellcheck disable=SC2206
+
+#SBATCH -A ADD_THE_PROJECT_NAME_HERE # project name 
+#SBATCH -M ADD_THE_SYSTEM_NAME_HERE # name of system i.e. snowy, dardel
+
+#SBATCH -p node # request a full node 
+#SBATCH -N 2  # Number of nodes  Change --num-gpus "2" in head command and worker loop as well
+#SBATCH -t 0:15:00 # change time accordingly
+#SBATCH --gpus-per-node=2  # change gpus accordingly
+#SBATCH -J exp-seed-3 # name of the job 
+#SBATCH -D ./ # stay in current working directory 
+
+
+source ~/.bashrc
+conda activate recover
+
+set -x
+
+# __doc_head_address_start__
+
+# Getting the node names
+nodes=$(scontrol show hostnames "$SLURM_JOB_NODELIST")
+nodes_array=($nodes)
+
+head_node=${nodes_array[0]}
+head_node_ip=$(srun --nodes=1 --ntasks=1 -w "$head_node" hostname --ip-address)
+
+# if we detect a space character in the head node IP, we'll
+# convert it to an ipv4 address. This step is optional.
+if [[ "$head_node_ip" == *" "* ]]; then
+IFS=' ' read -ra ADDR <<<"$head_node_ip"
+if [[ ${#ADDR[0]} -gt 16 ]]; then
+  head_node_ip=${ADDR[1]}
+else
+  head_node_ip=${ADDR[0]}
+fi
+echo "IPV6 address detected. We split the IPV4 address as $head_node_ip"
+fi
+# __doc_head_address_end__
+
+
+# __doc_head_ray_start__
+port=6379
+ip_head=$head_node_ip:$port
+export ip_head
+echo "IP Head: $ip_head"
+
+echo "Starting HEAD at $head_node"
+srun --nodes=1 --ntasks=1 -w "$head_node" \
+    ray start --head --node-ip-address="$head_node_ip" --port=$port \
+    --num-cpus "16" --num-gpus "2" --block &
+# __doc_head_ray_end__
+
+
+# __doc_worker_ray_start__
+# optional, though may be useful in certain versions of Ray < 1.0.
+sleep 10
+
+# number of nodes other than the head node
+worker_num=$((SLURM_JOB_NUM_NODES - 1))
+
+for ((i = 1; i <= worker_num; i++)); do
+    node_i=${nodes_array[$i]}
+    echo "Starting WORKER $i at $node_i"
+    srun --nodes=1 --ntasks=1 -w "$node_i" \
+        ray start --address "$ip_head" \
+        --num-cpus "16" --num-gpus "2" --block &
+    sleep 5
+done
+# __doc_worker_ray_end__
+
+
+# __doc_script_start__
+#Active learning with Upper Confidence Bound Aquisition 
+python train.py --config active_learning_UCB