InSpecLearn4SDL

Interpretable Spectral Learning for Self-Driving Labs

📘 Overview

InSpecLearn4SDL is the official implementation of the methods described in
📄 "Interpretable Spectral Features Predict Conductivity in Self-Driving Doped Conjugated Polymer Labs" (arXiv:2509.21330).

The repository provides an interpretable QSPR pipeline that predicts the electrical conductivity of doped conjugated polymers using optical spectra and processing parameters.
It is designed for integration into Self-Driving Labs (SDLs), enabling data-efficient, automated, and interpretable property prediction workflows.

---

🧠 Key Highlights

• 🔍 Spectral Featurization:
  Genetic Algorithm is used to adaptively select important spectral regions and use the area under the curve (AUC) as features

• 🧩 Interpretability:
  Domain-knowledge-driven feature expansion and SHAP-based feature selection retain physically meaningful descriptors.

• 📈 Performance:
  The hybrid model (expert + data-driven features) achieves high predictive accuracy while reducing experimental effort by ~33%.

• 🔬 Generalizable:
  Extendable to other spectroscopy–property relationships (e.g. Raman, FTIR, XANES).

---

🧪 Data Set — Experimental Data from the Amassian Group (North Carolina State University)

• 🧫 Processing Conditions:
  Variation in solvent concentration and annealing temperature across experiments.

• 🌈 Spectroscopic Measurements:
  Includes three spectral types —
  Pre-anneal UV–Vis, Post-anneal UV–Vis, and Post-dope UV–Vis–NIR spectra.

• 📊 Dataset Size:
  A total of 128 doped conjugated polymer samples, each with paired spectral and conductivity measurements.

---

🗂️ Project Structure

• Code/

  • main_final_QSPR_models.ipynb — 💡 Main Jupyter notebook containing the complete QSPR modeling workflow
  • helper.py — 🧰 Utility functions for data preprocessing and analysis
  • generate_adaptive_boundaries_optimization.py — 🧬 Genetic Algorithm for adaptive spectral boundary optimization
  • find_correct_clusters_and_do_ks_test.py — 📊 Train/test data clustering, KS-tests, and statistical validation

• Data/

  • spring24_solvtemp_jpm.csv — 📁 Main experimental dataset (solvent concentration and annealing temperature)
  • [experiment folders]/ — 🧪 Individual experiment spectral data files

---