AGENT SKILL

deepchem

Use DeepChem for molecular machine learning, data loading, featurization, model training, MoleculeNet, docking, and optional backend troubleshooting.

License 许可
MIT
最近更新
2026-09-03
Stars
★ 351
Forks
27

01 前言 Front Matter

name
deepchem
license
MIT
metadata
{ "disco-role": "operating" }
description
Use DeepChem for molecular machine learning, data loading, featurization, model training, MoleculeNet, docking, and optional backend troubleshooting.
disable-model-invocation
true

02 资源文件 5

03 SKILL.md 正文

DeepChem

Use this skill when a task involves the DeepChem Python package: molecular machine learning, drug-discovery or life-science datasets, molecule featurization, MoleculeNet loaders, model training/evaluation, structure/docking workflows, or optional TensorFlow/PyTorch/JAX/DQC dependency issues.

DeepChem is broad. Start here for routing, installation checks, and cross-cutting troubleshooting, then open the focused sub-skill for the workflow.

Quick Start

import deepchem as dc
print(dc.__version__)

For a minimal core install, expect numpy<2, pandas, scikit-learn, scipy>=1.10.1, sympy, joblib, and rdkit. Neural, docking, materials, and quantum workflows often need optional extras or external binaries.

Read references/installation.md when the request mentions installation, optional extras, backend warnings, CUDA, Colab, conda, or from-source setup. Read references/troubleshooting.md when imports succeed but DeepChem prints missing optional dependency warnings.

Route By Task

  • Data loading, MoleculeNet, splitters, transformers, CSV/SDF/FASTA/image/numpy datasets: use sub-skills/data-and-molnet/.
  • Featurizer selection, molecule fingerprints, graph features, RDKit descriptors, sequences, materials, polymers, or invalid SMILES diagnostics: use sub-skills/featurization/.
  • Model fitting, metrics, predictions, save/restore, callbacks, hyperparameter search, or model-family selection: use sub-skills/model-training/.
  • Protein-ligand structures, binding pockets, docking, PDB/PDBQT/SDF preparation, materials/DFT dependency triage, or external docking binaries: use sub-skills/docking-and-structure/.

If a user asks for an end-to-end molecular property workflow, combine the route order: data-and-molnet for loading/splitting, featurization for representation choice, then model-training for model/metric/evaluation. If a user asks for PDBBind, protein-ligand interactions, or docking scores, start in docking-and-structure, then route back to data/model sub-skills only after the structure-specific inputs are clear.

Common Workflows

  • Tiny molecular ML baseline: load a CSV with CSVLoader, choose CircularFingerprint or RDKitDescriptors, split with RandomSplitter or ScaffoldSplitter, train a SklearnModel, and evaluate with Metric.
  • MoleculeNet dataset: use dc.molnet.load_* only when downloads/cache are acceptable; otherwise build a small local fixture and follow data-and-molnet to avoid network surprises.
  • Graph neural model plan: use featurization to confirm MolGraphConvFeaturizer output and model-training to check whether torch/tensorflow/jax/DGL extras are installed before selecting a neural model.
  • Docking or pocket workflow: use docking-and-structure to check file formats and optional Vina/GNINA/OpenMM/PDBFixer/MDTraj dependencies before promising executable docking.

Bundled Checks

Run these skill-owned helpers from any environment where DeepChem is installed:

python sub-skills/data-and-molnet/scripts/load_tiny_csv_dataset.py
python sub-skills/data-and-molnet/scripts/split_tiny_dataset.py
python sub-skills/featurization/scripts/inspect_featurizer_outputs.py --smiles CCO c1ccccc1
python sub-skills/model-training/scripts/train_tiny_sklearn_model.py
python sub-skills/docking-and-structure/scripts/check_structure_dependencies.py

Use scripts/deepchem_environment_report.py for a cross-cutting import/version/optional dependency summary before debugging user environments.

Important References

  • references/installation.md: install modes, extras, and dependency expectations.
  • references/troubleshooting.md: cross-cutting import, optional dependency, data, and workflow failures.
  • references/repo-provenance.md: source snapshot and evidence paths used to generate this skill.

Boundaries

This skill is for using DeepChem as a package, not for maintaining DeepChem releases, CI, Docker images, or legacy contrib/ experiments. It intentionally treats large benchmarks, notebooks, remote dataset downloads, and GPU training as opt-in workflows rather than default verification paths.

04 同仓库更多 Skills

VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-triton-opt

Make a single PassNet Triton kernel fast AND numerically faithful: performance model (when a replacement can win at all), block/grid/warp tuning, autotune policy, launch overhead, and per-op numeric recipes to pass the dtype baseline tolerances. Use when a pass matches and is correct but speedup below expected, or when correctness fails by small numeric margins.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-solve

END-TO-END playbook for solving one PassNet sample: analyze the computation graphs, decide the optimization strategy (what to fuse, what to replace, what to leave alone), drive the iteration loop, and maximize the sample score. This is the ENTRY skill — invoke it first for any PassNet optimization task; it tells you when to use passnet-pattern-fusion, passnet-triton-opt and passnet-feedback.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-skill

PassNet GPU kernel optimization via compiler passes. Design and implement Triton-based optimization passes, create pass files under ./pass_dir/, self-evaluate with pass_evaluator, and iterate to maximize GPU speedup.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-pattern-fusion

Author PassNet pass files: write patterns that actually MATCH the FX graph, pick fusion regions (which ops to absorb into one Triton kernel), and structure multi-pass submissions with the mandatory shared-dispatch architecture. Use when creating or fixing pattern()/replacement_args()/replacement_func() files, when a pass "failed to match", or when deciding how to fuse multiple kernels/ops.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-orchestrate

Round/batch planning and decision support for PassNet graph optimization. Use for multi-sample triage, worker allocation, eval-budget policy, and post-eval keep/revert decisions. For solving one concrete sample end to end, use passnet-solve as the entry skill; consult this skill only when you need broader planning or a specific decision gate.

★ 351
VectorSpaceLab
VectorSpaceLab/AREX-Skill

passnet-feedback

Fast, deterministic feedback for PassNet work: pre-flight pattern/match verification WITHOUT burning a GPU evaluation, per-node bottleneck analysis of a sample's graphs, and eval-log parsing into per-variant status + estimated score + failure classification. Use BEFORE every GPU evaluation (check_pattern), at the START of a sample (analyze_graph), and AFTER every evaluation (parse_eval_log).

★ 351