$ show plinder-org/plinder
PLINDER is a comprehensive protein-ligand interaction dataset and evaluation resource, providing over 400,000 annotated systems with 750+ annotation columns derived from the PDB. The repository contains a full data generation pipeline (Metaflow-based, configurable via YAML) that downloads, annotates, scores, clusters, and splits structures into train/val/test sets with leakage control. It also includes a Python API (PlinderSystem, PlinderDataset) for accessing the dataset, querying similarity scores, and loading structures, as well as an evaluation harness (ModelScores, StratifiedTestSet) for benchmarking docking algorithms using metrics like lDDT-PLI, DockQ, and posebusters checks. The codebase is hybrid: primarily a data research project with strong software engineering practices (Docker, CI/CD, extensive tests, config-driven pipelines), but the core value is the curated dataset and the tools to generate and evaluate it.
The LLM identified 6 files worth highlighting for reuse:
A self-contained Pydantic model for protein-ligand complexes with methods for superposition, alignment, property extraction, and chain manipulation. Reusable in any structural biology project requiring a clean abstraction over mmCIF/PDB data.
Generates DuckDB SQL queries from pyarrow schemas and filter conditions, enabling efficient querying of large parquet datasets. The pattern is reusable for any project needing dynamic SQL generation from columnar data.
Implements a scatter-gather ETL pipeline with configurable stages (download, annotation, scoring, clustering, splitting) using OmegaConf. The pattern is reusable for any multi-stage data processing workflow that needs parallelism and checkpointing.
Performs graph-based train/val/test splitting with configurable similarity thresholds and leakage detection using networkit. Reusable for any ML project requiring stratified splits with controlled data leakage.
Provides ModelScores class that computes docking metrics (lDDT-PLI, BISY RMSD, posebusters checks) from predicted and reference structures. Reusable as a standalone evaluation harness for protein-ligand docking challenges.
Handles lazy download of large datasets from Cloudflare R2 with manifest-based integrity verification (size, MD5), offline caching, and retry logic. Reusable for any project needing reliable cloud data access with minimal local storage.