Abstract:It is a big challenge to develop efficient models for identifying personalized drug targets (PDTs) from high-dimensional personalized genomic profile of individual patients. Recent structural network control principles have introduced a new approach to discover PDTs by selecting an optimal set of driver genes in personalized gene interaction network (PGIN). However, most of current methods only focus on controlling the system through a minimum driver-node set and ignore the existence of multiple candidate driver-node sets for therapeutic drug target identification in PGIN. Therefore, this paper proposed multi-objective optimization-based structural network control principles (MONCP) by considering minimum driver nodes and maximum prior-known drug-target information. To solve MONCP, a discrete multi-objective optimization problem is formulated with many constrained variables, and a novel evolutionary optimization model called LSCV-MCEA was developed by adapting a multi-tasking framework and a rankings-based fitness function method. With genomics data of patients with breast or lung cancer from The Cancer Genome Atlas database, the effectiveness of LSCV-MCEA was validated. The experimental results indicated that compared with other advanced methods, LSCV-MCEA can more effectively identify PDTs with the highest Area Under the Curve score for predicting clinically annotated combinatorial drugs. Meanwhile, LSCV-MCEA can more effectively solve MONCP than other evolutionary optimization methods in terms of algorithm convergence and diversity. Particularly, LSCV-MCEA can efficiently detect disease signals for individual patients with BRCA cancer. The study results show that multi-objective optimization can solve structural network control principles effectively and offer a new perspective for understanding tumor heterogeneity in cancer precision medicine. The source code of our LSCV-MCEA and supplementary files can be freely downloaded from https://github.com/WilfongGuo/MONCP, with all data in this study.

Latent space search based multimodal optimization with personalized edge-network biomarker for multi-purpose early disease prediction

Multimodal Survival Ensemble Network: Integrating Genomic and Histopathological Insights for Enhanced Cancer Prognosis.

Multi-parameter systematic strategies for predictive, preventive and personalised medicine in cancer

Deep Biological Pathway Informed Pathology-Genomic Multimodal Survival Prediction

Deep Multi-path Network Integrating Incomplete Biomarker and Chest CT Data for Evaluating Lung Cancer Risk

MoME: Mixture of Multimodal Experts for Cancer Survival Prediction

Integrative data analysis of multi-platform cancer data with a multimodal deep learning approach

Dynamic Meta-data Network Sparse PCA for Cancer Subtype Biomarker Screening

Towards Non-invasive and Personalized Management of Breast Cancer Patients from Multiparametric MRI via A Large Mixture-of-Modality-Experts Model

Deep learning-based multi-modal data integration enhancing breast cancer disease-free survival prediction

Multi-task deep latent spaces for cancer survival and treatment response prediction

Multi-task deep latent spaces for cancer survival and drug sensitivity prediction

A multimodal ensemble approach for clear cell renal cell carcinoma treatment outcome prediction

Multi-Objective Optimization Based Network Control Principles for Identifying Personalized Drug Targets With Cancer

Identification of Pan-Cancer Prognostic Biomarkers Through Integration of Multi-Omics Data

Pancancer survival prediction using a deep learning architecture with multimodal representation and integration

SELECTOR: Heterogeneous graph network with convolutional masked autoencoder for multimodal robust prediction of cancer survival

Bi-level Graph Learning Unveils Prognosis-Relevant Tumor Microenvironment Patterns in Breast Multiplexed Digital Pathology

GD‐Net: An Integrated Multimodal Information Model Based on Deep Learning for Cancer Outcome Prediction and Informative Feature Selection