STRC Research

README

2 min read

Phase 5e-v2 ρ optimizer (autoresearch Pilot Zero)

This directory is an autoresearch-style closed-loop experiment harness.

One scalar: Spearman ρ(Vina mean ΔG, −τRAMD median) on the v5.2 shortlist. One file under iteration: ~/STRC/hypotheses/h01-pharmacochaperone/scripts/phase5e_v2_mutant_redock.py. One agent: read program.md, run python eval.py --experiment-id <slug>, observe, decide, repeat.

Files

  • program.md — the full instruction set (read first)
  • eval.py — runs the orchestrator on v5.2-shortlist + computes ρ + appends to history.jsonl (DO NOT MODIFY)
  • history.jsonl — append-only experiment log
  • summary.md — written at end-of-run by the agent
  • notes.md — agent’s scratch space (optional)

Quick start (human-driven)

cd ~/STRC/autoresearch/phase5e_v2_rho_optimizer
python eval.py --experiment-id baseline_2026-04-26 --diff-summary "baseline reproduction"
# Wait ~17 min, read history.jsonl, decide

Agent-driven (overnight)

Spin up Claude/Codex with permissions scoped to:

  • This directory (read+write)
  • ~/STRC/hypotheses/h01-pharmacochaperone/scripts/phase5e_v2_mutant_redock.py (write)
  • ~/STRC/hypotheses/h01-pharmacochaperone/artifacts/phase5e_v2/v5_2-shortlist/ (write+delete)
  • pip install and shell-out to obabel / antechamber / prepare_flexreceptor.py / mk_prepare_ligand.py

Then prompt: “Read program.md and run experiment 1.”

Bridge to Brain memory / NGAM

Each history.jsonl line is a structured fact about Vina rigid-receptor behavior on anion-leads. NGAM (Brain connection-making project, in flight 2026-04-26) can ingest this log and surface relevant past experiments to new optimizer pilots (Phase 5q v5.3, Phase 4f MM-GBSA, etc.).

Status

Pilot Zero — not yet run. Scaffold ready 2026-04-26. First experiment pending agent spin-up.

Graph View