# Integrated bioinformatic analysis and machine learning developed a prognostic model based on mitochondrial function for acute myeloid leukemia

## TL;DR

This study uses mitochondrial gene data and machine learning to create a prognostic model for acute myeloid leukemia, helping predict patient outcomes and treatment responses.

## Contribution

A novel MitoScore model was developed by integrating mitochondrial gene expression and machine learning for AML prognosis.

## Key findings

- High-risk MitoScore groups showed worse prognosis and enriched immune-related pathways.
- MitoScore identified gene mutations and immune infiltration differences between risk groups.
- The model demonstrated greater sensitivity to specific drugs in high-risk groups.

## Abstract

The disease burden of acute myeloid leukemia (AML) continues to pose a significant public health challenge globally. Mitochondria play a critical role in tumor development and progression by influencing bioenergetics, biosynthesis, and signaling pathways. However, the prognostic significance and therapeutic implications of mitochondrial function in AML warrant further investigation.

We integrated mitochondrial gene expression data with bulk RNA sequencing to identify key mitochondrial genes associated with AML. A total of fourteen machine learning algorithms were employed, yielding 148 unique combinations. The best-performing model was utilized to develop a MitoScore, which was then combined with clinical variables to establish a MitoScore-based nomogram. Additionally, single-cell sequencing data were analyzed to assess the impact of key mitochondrial genes on immune cells. Samples were classified into low-risk and high-risk groups based on MitoScore, allowing for a comparative analysis of clinical features, biological mechanisms, copy number variations, tumor burden, immune infiltration, immune function, and drug sensitivity between the two groups.

Specific expression patterns of mitochondrial genes were observed in T cell subsets and at various developmental stages of AML. Samples were classified into low-risk and high-risk groups based on MitoScore. The high-risk MitoScore group exhibited a worse prognosis, with enriched biological processes and molecular pathways associated with immune response, a higher frequency of gene mutations linked to poor outcomes, increased immune cell infiltration, enhanced immune function, upregulated immune checkpoint gene expression, and greater sensitivity to cyclophosphamide and venetoclax.

This robust machine learning framework underscores the potential of MitoScore as a tool for stratified prognostic assessment and personalized treatment planning in AML patients.