From f0f9d9e1a750d16f98e0e70b1277185933f4dc12 Mon Sep 17 00:00:00 2001 From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com> Date: Mon, 26 Jan 2026 12:53:25 +0000 Subject: [PATCH 1/5] Initial plan From 873902a13b56c1c10a621c5b7976cf08a6409e96 Mon Sep 17 00:00:00 2001 From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com> Date: Mon, 26 Jan 2026 12:56:57 +0000 Subject: [PATCH 2/5] Add concise quick usage summary to README Co-authored-by: thkim-01 <248642018+thkim-01@users.noreply.github.com> --- README.md | 5 +++++ 1 file changed, 5 insertions(+) diff --git a/README.md b/README.md index ac71954..b9c2121 100644 --- a/README.md +++ b/README.md @@ -8,6 +8,11 @@ DRGS-Net implements a hybrid molecular representation model that concatenates: The combined embedding is passed to a small prediction head for downstream molecular property prediction (classification or regression). +## TL;DR (Quick usage) +1) Copy the default config: `cp DRGS-Net.yaml config_concatenate.yaml` +2) Edit `config_concatenate.yaml` with your dataset paths and ChemBERTa checkpoint. +3) Run: `python DRGS-Net_finetune.py` + This repository contains training/finetuning scripts, dataset wrappers, model definitions, and utilities. The hybrid model is implemented in `models/concatenate_model.py` (class `HybridModel`). The finetuning orchestration is in `DRGS-Net_finetune.py` which expects a configuration file named `config_concatenate.yaml` by default — if you only have `DRGS-Net.yaml`, copy/rename it to `config_concatenate.yaml` before running (see "Quick start"). ## Methodology ![DRGS-Net Architecture](./fig/Methodology.png) From c4357366d29b7f8c89b760d6657a68fa24a72917 Mon Sep 17 00:00:00 2001 From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com> Date: Mon, 26 Jan 2026 12:57:34 +0000 Subject: [PATCH 3/5] Clarify quick usage dependencies in README Co-authored-by: thkim-01 <248642018+thkim-01@users.noreply.github.com> --- README.md | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/README.md b/README.md index b9c2121..bf76412 100644 --- a/README.md +++ b/README.md @@ -9,9 +9,10 @@ DRGS-Net implements a hybrid molecular representation model that concatenates: The combined embedding is passed to a small prediction head for downstream molecular property prediction (classification or regression). ## TL;DR (Quick usage) -1) Copy the default config: `cp DRGS-Net.yaml config_concatenate.yaml` -2) Edit `config_concatenate.yaml` with your dataset paths and ChemBERTa checkpoint. -3) Run: `python DRGS-Net_finetune.py` +1) Install dependencies (see "Quick start" below for full setup). +2) Copy the default config: `cp DRGS-Net.yaml config_concatenate.yaml` +3) Edit `config_concatenate.yaml` with your dataset paths and ChemBERTa checkpoint. +4) Run: `python DRGS-Net_finetune.py` This repository contains training/finetuning scripts, dataset wrappers, model definitions, and utilities. The hybrid model is implemented in `models/concatenate_model.py` (class `HybridModel`). The finetuning orchestration is in `DRGS-Net_finetune.py` which expects a configuration file named `config_concatenate.yaml` by default — if you only have `DRGS-Net.yaml`, copy/rename it to `config_concatenate.yaml` before running (see "Quick start"). ## Methodology From 11ea8638061d52f6bf719c3a6d57d915d665f1da Mon Sep 17 00:00:00 2001 From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com> Date: Mon, 26 Jan 2026 12:58:06 +0000 Subject: [PATCH 4/5] Refine quick start reference in README Co-authored-by: thkim-01 <248642018+thkim-01@users.noreply.github.com> --- README.md | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/README.md b/README.md index bf76412..c9b1cdd 100644 --- a/README.md +++ b/README.md @@ -9,7 +9,7 @@ DRGS-Net implements a hybrid molecular representation model that concatenates: The combined embedding is passed to a small prediction head for downstream molecular property prediction (classification or regression). ## TL;DR (Quick usage) -1) Install dependencies (see "Quick start" below for full setup). +1) Install dependencies (see "Quick start" below for full setup commands). 2) Copy the default config: `cp DRGS-Net.yaml config_concatenate.yaml` 3) Edit `config_concatenate.yaml` with your dataset paths and ChemBERTa checkpoint. 4) Run: `python DRGS-Net_finetune.py` From 281fbf682e4a096e4c4f9082a03dd62d2a8d47cc Mon Sep 17 00:00:00 2001 From: "copilot-swe-agent[bot]" <198982749+Copilot@users.noreply.github.com> Date: Mon, 26 Jan 2026 13:01:56 +0000 Subject: [PATCH 5/5] Add repo download step to Quick start Co-authored-by: thkim-01 <248642018+thkim-01@users.noreply.github.com> --- README.md | 15 +++++++++++---- 1 file changed, 11 insertions(+), 4 deletions(-) diff --git a/README.md b/README.md index c9b1cdd..feb0f90 100644 --- a/README.md +++ b/README.md @@ -56,7 +56,14 @@ See the model card: https://huggingface.co/DeepChem/ChemBERTa-77M-MLM ## Quick start -1) Prepare config file +1) Download the repository + +```bash +git clone https://github.com/thkim-01/DRGS-Net.git +cd DRGS-Net +``` + +2) Prepare config file The finetune script (`DRGS-Net_finetune.py`) by default loads `config_concatenate.yaml`. If you only have `DRGS-Net.yaml`, create a copy with the expected name: @@ -66,7 +73,7 @@ cp DRGS-Net.yaml config_concatenate.yaml Edit the YAML to set `task_name`, dataset paths, `fine_tune_from` (pretrained GNN checkpoint folder under `./ckpt/`) and `hybrid_specific.chemberta_model_name` (local path or Hugging Face model id). -2) Create environment & install dependencies (suggested) +3) Create environment & install dependencies (suggested) This project requires PyTorch, HuggingFace Transformers, and optional packages like RDKit and NVIDIA Apex for mixed precision. @@ -86,11 +93,11 @@ pip install tensorboard scikit-learn pandas numpy tqdm pyyaml Note: Install compatible `torch-geometric` packages for your PyTorch/CUDA setup if you use PyG layers in the GNN models. -3) Prepare data +4) Prepare data Place downstream datasets under `data//` following the MoleculeNet CSV formats. The default config uses paths like `data/bbbp/BBBP.csv` etc. See `DRGS-Net_finetune.py` for dataset mapping by `task_name`. -4) Run finetuning +5) Run finetuning ```bash python DRGS-Net_finetune.py