MulMarker : an effective tool for identifying multi-gene prognostic signatures

Abstract

Personalized medicine is increasingly shaping the future of healthcare, allowing clinicians to tailor medical treatments to distinct subsets of patients, further enhancing therapeutic efficacy based on individual’s characteristics. Prognostic signatures can serve as selection criteria in clinical trials to pinpoint patients at higher risk for adverse clinical outcomes or disease progression, thus widely used in patient care, clinical trials, and drug development. With the advent of next-generation sequencing, vast amounts of sequencing data are generated, leading to a growing trend in using computational methods to identify prognostic signatures. While various tools exist for identifying single-gene prognostic biomarkers, there is a notable gap in tools for multi-gene prognostic signature identification. Compared to single-gene biomarkers, multi-gene prognostic signatures provide a deeper understanding of disease mechanics, revealing complex genetic interplays significant for disease progression and treatment response. This advanced insight is crucial for accurate patient risk assessment and tailored therapeutic strategies, underscoring their essential role in personalized medicine.

To address the gap, we propose MulMarker, an effective tool designed to identify multi-gene prognostic signatures. By integrating data preprocessing, feature selection, model construction, and prognostic signature evaluation into the tool, MulMarker provides a robust and efficient solution for the identification of multigene prognostic signatures. Central to its efficiency, MulMarker consists of three core modules: (1) a chatbot to answer user queries about the inputs, algorithms, and analysis details; (2) a module for identifying multi-gene prognostic signatures; and (3) a module for generating tailored reports to provide users with comprehensive and understandable results. MulMarker is highlighted by several key advantages. Firstly, it provides a solution to identify multi-gene prognostic signatures. By employing multivariate Cox regression analysis to combine multiple genes as a prognostic signature, MulMarker enables to analyze the combined effect of multiple genes on patient outcomes, such as survival or disease progression. Second, MulMarker provides a comprehensive system spanning the entire process from identifying to evaluating prognostic signatures. Besides, we integrated large language models into the tool for better addressing users queries and generating tailored reports, providing an immediate response and user-friendly solution. In terms of robustness, MulMarker is underscored through a case study on breast cancer, where it successfully identifies a cell-cycle-related prognostic signature, which is validated experimentally. Additionally, the versality of MulMarker was highlighted by its application in identifying a genome-instability prognostic signature for lung adenocarcinoma, revealing how MulMarker can be adapted to integrate disease-specific biological characteristic for the identification of prognostic signatures. Notably, the identified prognostic signature for lung adenocarcinoma outperformed existing state-of-the-art prognostic signatures, underscoring the potential of the tool. Moreover, we have expanded the usage of MulMarker to various diseases, only if the input data is quantitative values. Overall, MulMarker presents a promising solution to the challenges in identifying multi-gene prognostic signatures and realizing their potential in clinical practice.