coreason-prism 0.3.0

coreason-prism

coreason-prism

The Scientific Eye / Multi-Modal Encoder

coreason-prism is the specialized processing engine for scientific data types (Chemistry and Vision) within the CoReason AI ecosystem. It acts as the "Scientific Eye" and "Multi-Modal Encoder", transforming fragile string representations and static images into robust, mathematical graphs and vectors.

Core Philosophy:

"A Molecule is a Graph, not a String. A Chart is Data, not an Image."

---

Features

Derived from the Product Requirements Document:

• Cheminformatic Grounding (The Chemist):

  • Treats molecules as mathematical graphs, not strings.
  • Normalizes and sanitizes chemical structures (SMILES/InChI) using datamol.
  • Transmutes SMILES to SELFIES for 100% valid generative output.
  • Computes fingerprints (Morgan/ECFP) for structural similarity search.
  • Calculates key properties: Molecular Weight, LogP, TPSA, Lipinski Violations.

• Visual De-Plotting (The Analyst):

  • Extracts raw data from scientific figures (e.g., Kaplan-Meier curves) using DePlot.
  • Digitizes charts into linear tables/DataFrames.
  • Enables meta-analysis of data locked in PDF images.

• Bio-Image Segmentation (The Biologist):

  • Segments and classifies medical images (e.g., Histology) using MedSAM.
  • Detects ROIs and computes metrics like cell counts and tumor area.

• Multi-Modal Embedding (The Embedder):

  • Generates joint embeddings for text, molecules, and images using BioCLIP.
  • Enables multi-modal retrieval (e.g., searching for histology slides via text description).

---

Installation

pip install coreason-prism

Or install from source:

git clone https://github.com/CoReason-AI/coreason_prism.git
cd coreason_prism
pip install .

Usage

Here is a concise snippet showing how to initialize and use the library:

from pathlib import Path
from coreason_prism.interface import Prism, PrismMode

# Initialize Prism (The Facade)
prism = Prism(light_mode=False)  # Set light_mode=True to skip heavy models (DePlot/BioCLIP)

# 1. Process a Molecule (SMILES -> Graph/SELFIES + Properties)
molecule_result = prism.process_molecule("CC(=O)Oc1ccccc1C(=O)O")  # Aspirin
if molecule_result.status == "VALID":
    print(f"Canonical SMILES: {molecule_result.canonical_smiles}")
    print(f"SELFIES: {molecule_result.selfies_string}")
    print(f"LogP: {molecule_result.logp}")
    print(f"Fingerprint (first 10 bits): {molecule_result.fingerprint_vector[:10]}")

# 2. Process a Chart Image (Extract Data)
# Ensure you have an image file at the specified path
chart_path = Path("tests/data/kaplan_meier.png")
if chart_path.exists():
    chart_result = prism.process_image(
        image_path=chart_path,
        source_document_id="doc_123",
        mode=PrismMode.CHART
    )
    print(f"Figure Type: {chart_result.figure_type}")
    print(f"Extracted Data: {chart_result.data_series}")
    if chart_result.metadata:
        print(f"Median Survival: {chart_result.metadata.get('median_survival')}")

# 3. Process a Bio-Image (Segmentation)
bio_path = Path("tests/data/histology_slide.jpg")
if bio_path.exists():
    bio_result = prism.process_image(
        image_path=bio_path,
        source_document_id="doc_456",
        mode=PrismMode.BIO
    )
    print(f"Cell Count: {bio_result.metadata.get('cell_count')}")