Autoencoder-based multimodal prediction of non-small cell lung cancer survival

Abstract

The ability to accurately predict non-small cell lung cancer (NSCLC) patient survival is crucial for informing physician decision-making, and the increasing availability of multi-omics data offers the promise of enhancing prognosis predictions. We present a multimodal integration approach that leverages microRNA, mRNA, DNA methylation, long non-coding RNA (lncRNA) and clinical data to predict NSCLC survival and identify patient subtypes, utilizing denoising autoencoders for data compression and integration. Survival performance for patients with lung adenocarcinoma (LUAD) and squamous cell carcinoma (LUSC) was compared across modality combinations and data integration methods. Using The Cancer Genome Atlas data, our results demonstrate that survival prediction models combining multiple modalities outperform single modality models. The highest performance was achieved with a combination of only two modalities, lncRNA and clinical, at concordance indices (C-indices) of 0.69 ± 0.03 for LUAD and 0.62 ± 0.03 for LUSC. Models utilizing all five modalities achieved mean C-indices of 0.67 ± 0.04 and 0.63 ± 0.02 for LUAD and LUSC, respectively, while the best individual modality performance reached C-indices of 0.64 ± 0.03 for LUAD and 0.59 ± 0.03 for LUSC. Analysis of biological differences revealed two distinct survival subtypes with over 900 differentially expressed transcripts.

Survival Prediction Landscape: An In-Depth Systematic Literature Review on Activities, Methods, Tools, Diseases, and Databases

“…The specifics of different predictors, in terms of variations in the combinations of clinical and various omics data modalities, are outlined in Table 4. Among 54 survival prediction studies based on multiomics, 49 studies utilized different combinations of four distinct omics types: mRNA, methylation, miRNA, and CNV 14, 26, 27, 42, 43, 69, 72, 73, 77, 82, 84, 89, 96, 97, 100, 101, 106, 108 . Only 7 studies utilized additional modalities such as whole exome sequencing (WES) 26,31 , long coding RNA (lncRNA) 31 , proteomics 22, 23, 108, 113, 115 , and mutation data 22, 23, 108, 115 .…”

“…Among the 74 distinct survival prediction studies, only 26 have provided publicly accessible source code. Among these studies, 6 studies have utilized R 91,94,96,103,109,119 and 20 have opted for Python 14, 24, 28, 38, 72, 74, 82, 83, 89, 90, 95, 100, 105, 111, 116, 118, 131 23, 27, 106 . A comprehensive analysis of open source codes reveals that a majority of these tools have been developed from scratch without utilizing any specific survival prediction library 14, 28, 83, 95 .…”

Survival prediction landscape: an in-depth systematic literature review on activities, methods, tools, diseases, and databases

“…The specifics of different predictors, in terms of variations in the combinations of clinical and various omics data modalities, are outlined in Table 5 . Among 55 survival prediction studies based on multiomics, 49 studies utilized different combinations of four distinct omics types: mRNA, methylation, miRNA, and CNV (Baek and Lee, 2020 ; Jiang et al, 2020 ; Li et al, 2020 ; Tan et al, 2020 ; Tong D. et al, 2020 ; Tong L. et al, 2020 ; Yang Q. et al, 2020 ; Chai et al, 2021a ; Hira et al, 2021 ; Hu Q. et al, 2021 ; Owens et al, 2021 ; Tong et al, 2021 ; Zhang X. et al, 2021 ; Zhang Z.-S. et al, 2021 ; Zhao L. et al, 2021 ; Bhat and Hashmy, 2023 ; Ellen et al, 2023 ; Hao et al, 2023 ). Only seven studies utilized additional modalities such as whole exome sequencing (WES) (Baek and Lee, 2020 ; Jiang et al, 2022 ), long coding RNA (lncRNA) (Jiang et al, 2022 ), proteomics (Tan et al, 2020 ; Malik et al, 2021 ; Unterhuber et al, 2021 ; Richard et al, 2022 ; Pellegrini, 2023 ), and mutation data (Tan et al, 2020 ; Malik et al, 2021 ; Unterhuber et al, 2021 ; Pellegrini, 2023 ).…”

“…Among the 90 distinct survival prediction studies, only 28 have provided publicly accessible source code. Among these studies, six studies have utilized R (Kantidakis et al, 2020 ; Li et al, 2021 ; Redekar et al, 2022 ; Zhang S. et al, 2022 ; Ellen et al, 2023 ; Willems et al, 2023 ) and 22 have opted for Python (Jiang et al, 2020 ; Tong L. et al, 2020 ; Chai et al, 2021a ; Hathaway Q. A. et al, 2021 ; Hira et al, 2021 ; Malik et al, 2021 ; Poirion et al, 2021 ; Xu et al, 2021 ; Zhang X. et al, 2021 ; Zhao L. et al, 2021 ; Wang et al, 2022 ; Wu and Fang, 2022 ; Yin et al, 2022 ; Zhang J.…”