TILDA-X: Transcriptome-Informed Lung Cancer Disparities via Explainable AI.

Area of Science:

• Computational biology
• Genomics
• Artificial intelligence in medicine

Background:

• Lung cancer exhibits significant disparities in incidence and outcomes across racial and sex groups.
• Existing lung cancer datasets are imbalanced by race, potentially biasing disparity analyses.
• Identifying patient- and cohort-specific biomarkers is crucial for targeted lung cancer therapies.

Purpose of the Study:

• To develop an explainable AI framework (TILDA-X) for analyzing lung cancer disparities.
• To mitigate racial imbalance in datasets by classifying based on disease conditions rather than race.
• To delineate patient-specific and cohort-specific disparity information using transcriptome data.

Main Methods:

• Developed TILDA-X, an explainable AI framework for classification based on disease conditions (lung adenocarcinoma, lung squamous cell carcinoma, healthy).
• Utilized a lung cancer transcriptome dataset for model development.
• Employed a bottom-up approach to identify cohort-specific disparity information for various racial and sex groups.

Main Results:

• Disease condition-based classification achieved high accuracy (88-100%) for minority groups, outperforming race-based classification (0-16%).
• Functional analysis identified unique pathways associated with lung cancer in different racial and sex subgroups.
• Over 63% of identified pathways overlapped with existing lung cancer research, validating the findings.

Conclusions:

• The TILDA-X framework provides a robust and interpretable method for characterizing race- and sex-specific lung cancer disparities.
• This approach supports precision oncology by enabling the development of equitable therapies based on transcriptome profiles.
• The study highlights the importance of disease-condition-based AI models for addressing data imbalance and revealing hidden disparities.