Multi-Modal Deep Learning for COVID-19 Severity Assessment via Gene Expression and Clinical Data Integration
0
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
Accurately predicting the severity of severe acute respiratory syndrome (SARS-CoV-2) infection is crucial for effective patientmanagement and resource allocation. Building on the ExPDrug model, which utilizes gene expression data and gene-pathway relationshipsfor severity prediction, we develop a novel multi-modal model that integrates an additional modality by incorporating a projection layerto align heterogeneous data sources including clinical patient information. The experimental results demonstrate that the multimodalmodel outperforms ExPDrug in terms of the area under the curve (AUC), accuracy, and precision, particularly for CD8+ T cells. To enhancethe interpretability of the findings, the Local Interpretable Model-agnostic Explanations (LIME) method was employed, which identifiedrespiration rate and SpO2 as pivotal clinical indicators. These features may help explain the model’s strong performance for CD8+ Tcells, based on their relationship with specific immune cell activities. These observations underscore the synergistic value of integratinggene expression and clinical data for severity prediction in COVID-19.
Abstract
Accurately predicting the severity of severe acute respiratory syndrome (SARS-CoV-2) infection is crucial for effective patientmanagement and resource allocation. Building on the ExPDrug model, which utilizes gene expression data and gene-pathway relationshipsfor severity prediction, we develop a novel multi-modal model that integrates an additional modality by incorporating a projection layerto align heterogeneous data sources including clinical patient information. The experimental results demonstrate that the multimodalmodel outperforms ExPDrug in terms of the area under the curve (AUC), accuracy, and precision, particularly for CD8+ T cells. To enhancethe interpretability of the findings, the Local Interpretable Model-agnostic Explanations (LIME) method was employed, which identifiedrespiration rate and SpO2 as pivotal clinical indicators. These features may help explain the model’s strong performance for CD8+ Tcells, based on their relationship with specific immune cell activities. These observations underscore the synergistic value of integratinggene expression and clinical data for severity prediction in COVID-19.