Mdbabumiamssm Llms Universal Life Science And Clinical Skills Chic Ml Framework Agent

name: 'chic-ml-framework-agent' description: 'Machine learning framework for inferring high-risk clonal hematopoiesis from complete blood count data without sequencing, reducing the number needed to sequence for CHIP screening.' measurable_outcome: Execute skill workflow successfully with valid output within 15 minutes. allowed-tools: - read_file - run_shell_command

CHIC ML Framework Agent

The CHIC (Clonal Hematopoiesis Inference from Counts) ML Framework Agent uses machine learning to identify individuals with high-risk clonal hematopoiesis from routine complete blood count (CBC) data alone. Validated on 431,531 UK Biobank participants, CHIC reduces the "number needed to sequence" from 727 to 40 individuals per case of high-risk CH, enabling population-scale CHIP screening without universal sequencing.

When to Use This Skill

• When screening large populations for CHIP without sequencing.
• For prioritizing individuals who should undergo genetic testing.
• To identify undiagnosed CCUS/MDS from routine blood counts.
• When sequencing resources are limited but CHIP detection is needed.
• For research studies requiring CHIP prevalence estimation.

Core Capabilities

1. CBC-Based CHIP Prediction: Predict CHIP presence from blood counts.

2. High-Risk CH Identification: Focus on clinically relevant clones.

3. Efficient Screening: Dramatically reduce number needed to sequence.

4. CCUS/MDS Detection: Flag potential undiagnosed blood cancers.

5. Population Stratification: Risk-stratify for targeted sequencing.

6. Longitudinal Monitoring: Track CBC changes suggesting CH development.

CHIC Model Features

Performance Metrics

Workflow

1. Input: CBC data (complete blood count panel).

2. Feature Engineering: Calculate derived indices.

3. Model Inference: Run CHIC tree-based ensemble.

4. Risk Scoring: Generate CHIP probability score.

5. Stratification: Categorize into risk groups.

6. Recommendations: Prioritize for sequencing.

7. Output: Risk scores, sequencing prioritization list.

Example Usage

User: "Screen this population cohort using CHIC to identify individuals who should undergo CHIP sequencing."

Agent Action:

python3 Skills/Hematology/CHIC_ML_Framework_Agent/chic_screening.py \
    --cbc_data population_cbc.csv \
    --demographics demographics.csv \
    --model_weights chic_xgboost_ukbb.pt \
    --threshold stringent \
    --output_prioritization true \
    --output chic_screening_results/

Input Data Format

sample_id,age,sex,rbc,hgb,hct,mcv,mch,mchc,rdw,wbc,plt,mpv
001,68,M,4.2,13.5,40.2,95.7,32.1,33.6,14.8,7.2,245,10.2
002,72,F,3.8,11.2,34.5,90.8,29.5,32.5,16.2,8.5,312,9.8

Output Components

Risk Stratification

Model Architecture

AI/ML Components

Feature Engineering:

• Derived indices (RDW/MCV ratio, etc.)
• Age-sex adjusted values
• Longitudinal trajectory features

Model Training:

• UK Biobank 431,531 participants
• 20,860 sequencing-confirmed CH
• Tree-based ensemble methods

Threshold Optimization:

• Sensitivity-specificity tradeoffs
• NNS optimization
• Cost-effectiveness modeling

Clinical Applications

Validation Results

Prerequisites

• Python 3.10+
• XGBoost, scikit-learn
• pandas, numpy
• CHIC model weights
• CBC reference ranges

Related Skills

• CHIP_Clonal_Hematopoiesis_Agent - Full CHIP analysis with sequencing
• MPN_Progression_Monitor_Agent - MPN monitoring
• Blood_Smear_AI_Agent - Morphology analysis
• MDS_Classification_Agent - MDS diagnosis

CBC Red Flags for CH

Limitations

Special Considerations

1. Population Calibration: Model trained on UK Biobank demographics
2. Acute Illness: Exclude acutely ill patients
3. Recent Transfusion: Affects CBC parameters
4. Medications: Consider drug effects on CBC
5. Serial Monitoring: Longitudinal changes informative

Cost-Effectiveness

Future Directions

Author

AI Group - Biomedical AI Platform

Source: mdbabumiamssm/LLMs-Universal-Life-Science-and-Clinical-Skills- — distributed by TomeVault.