DeepQA: A Unified Transcriptome-Based Aging Clock Using Deep Neural Networks.

Area of Science:

• Genomics
• Computational Biology
• Biotechnology

Background:

• Predicting biological age from gene expression is crucial for understanding aging and developing therapeutics.
• Existing aging clocks often exhibit bias by training solely on healthy subjects, leading to inaccurate predictions for unhealthy individuals.
• Current transcriptome-based aging clocks utilize inefficient gene selection and conventional machine learning models.

Purpose of the Study:

• To develop a more accurate and less biased method for biological age estimation using gene expression data.
• To address the limitations of existing aging clocks, including bias and inefficient gene selection processes.
• To introduce DeepQA, a unified aging clock based on a mixture of experts approach.

Main Methods:

• Proposed DeepQA, a novel aging clock employing a mixture of experts architecture.
• Developed a specialized Hinge-Mean-Absolute-Error (Hinge-MAE) loss function for training on both healthy and unhealthy subjects across multiple cohorts.
• Implemented an approach that avoids inefficient exhaustive gene selection procedures.

Main Results:

• DeepQA demonstrated significantly superior performance in biological age estimation compared to existing methods.
• The proposed method effectively reduced bias when inferring biological age in unhealthy subjects.
• DeepQA provides a novel method for identifying genes involved in aging prediction, offering an alternative to traditional differential gene expression analysis.

Conclusions:

• DeepQA offers a more accurate and robust approach to biological age prediction from gene expression data.
• The method's ability to train on diverse cohorts mitigates bias, improving applicability to broader populations.
• DeepQA advances the field by providing an efficient and effective tool for aging research and biomarker discovery.