MOGAT: A Multi-Omics Integration Framework Using Graph Attention Networks for Cancer Subtype Prediction
Raihanul Bari Tanvir, Md Mezbahul Islam, Masrur Sobhan, Dongsheng Luo * and Ananda M. Mondal * Knight Foundation School of Computing and Information Sciences, Florida International University, Miami, FL 33199, USA; rtanv003@fiu.edu (R.B.T); misla093@fiu.edu (M.M.I); msobh002@fiu.edu (M.S.)
- Correspondence: dluo@fiu.edu (D.L.); amondal@fiu.edu (A.M.M.)
The salient features of this study are enumerated below.
• Our group is the first to explore graph attention network-based multi-omics inte-gration for cancer subtype prediction.
• The proposed approach, MOGAT, provides better embeddings than MOGONET and SUPREME for multi-omics integration, which resulted in improved accuracy for cancer subtype prediction.
• MOGAT embeddings provide better prognosis in differentiating the high-risk group from the low-risk group, which will help the physician to device appropriate treatment strategy for an individual patient depending on the location of the patient on the prognostic curve.
• Our group is the first to incorporate lncRNA expression in multi-omics integration studies.
• We provided detail information so that the results can be reproduced, such as (a) handling duplicate samples coming from the same patient and (b) providing the number of features in each step of preprocessing from raw features to cleaned fea-tures to selected features.
• The interactions between different omics types are considered during the node feature engineering by concatenating features from different omics types.
![image](https://private-user-images.githubusercontent.com/157138926/298217404-5d6f67c9-55c8-4358-ac93-bd9b4a13415c.png?jwt=eyJhbGciOiJIUzI1NiIsInR5cCI6IkpXVCJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3MjYwMTk1MzcsIm5iZiI6MTcyNjAxOTIzNywicGF0aCI6Ii8xNTcxMzg5MjYvMjk4MjE3NDA0LTVkNmY2N2M5LTU1YzgtNDM1OC1hYzkzLWJkOWI0YTEzNDE1Yy5wbmc_WC1BbXotQWxnb3JpdGhtPUFXUzQtSE1BQy1TSEEyNTYmWC1BbXotQ3JlZGVudGlhbD1BS0lBVkNPRFlMU0E1M1BRSzRaQSUyRjIwMjQwOTExJTJGdXMtZWFzdC0xJTJGczMlMkZhd3M0X3JlcXVlc3QmWC1BbXotRGF0ZT0yMDI0MDkxMVQwMTQ3MTdaJlgtQW16LUV4cGlyZXM9MzAwJlgtQW16LVNpZ25hdHVyZT03NjM0YTk2ZDI3YWFlNDAzOWViY2FlYTZmZWQxNGYxM2MzYjY0MzFhYTI0ZDc2MzEzY2IzYjBiZjllNDcyMDk4JlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCZhY3Rvcl9pZD0wJmtleV9pZD0wJnJlcG9faWQ9MCJ9.IXEs9BC2M7MWsyDNtsq1RENz_vxsa22mTJUh3Jr6cHo)
Figure: Illustration of MOGAT framework.