MTM

MTM predicts tissue-specific gene expression across multiple tissues using a unified deep learning multi-task framework and optimizes genetic manipulation schedules for strain development via Greedy Search of Common Ancestor Strains (GSCAS) and the Minimizing Total Manipulations (MTM) algorithm.

Key Features:

• Unified Deep Learning Framework: A unified deep learning model integrates multi-task learning to predict gene expression profiles across various tissues from reference samples and captures tissue-shared intrinsic relevance to enhance predictive accuracy.
• Individualized Cross-Tissue Predictions: Utilizes individualized cross-tissue information from reference samples to preserve biological variation and achieve superior performance at both the sample and gene levels for unseen individuals.
• Prediction Accuracy: Produces high-accuracy predictions of tissue gene expression suitable for downstream transcriptomic analyses.
• GSCAS (Greedy Search of Common Ancestor Strains): An algorithm that identifies common ancestor strains to reduce required manipulations when constructing multiple strains.
• Minimizing Total Manipulations (MTM) algorithm: An algorithm that minimizes the total number of gene manipulations needed across strain development workflows to reduce time and cost.

Scientific Applications:

• Cross-tissue transcriptomic inference: Predicts gene expression profiles for tissues that are difficult or impractical to sample directly.
• Fundamental and clinical research: Supports basic and clinical studies by broadening access to tissue-specific transcriptomic data without additional sampling.
• Biofoundry strain development optimization: Optimizes genetic manipulation schedules to reduce time and cost when constructing large numbers of strains using GSCAS and MTM algorithms.

Methodology:

Employs a unified deep learning model with multi-task learning that leverages individualized cross-tissue information from reference samples for expression prediction, and applies GSCAS and Minimizing Total Manipulations algorithms to optimize genetic manipulation schedules.