Patient similarity by joint matrix trifactorization to identify subgroups in acute myeloid leukemia

Abstract

Objective Computing patients’ similarity is of great interest in precision oncology since it supports clustering and subgroup identification, eventually leading to tailored therapies. The availability of large amounts of biomedical data, characterized by large feature sets and sparse content, motivates the development of new methods to compute patient similarities able to fuse heterogeneous data sources with the available knowledge. Materials and Methods In this work, we developed a data integration approach based on matrix trifactorization to compute patient similarities by integrating several sources of data and knowledge. We assess the accuracy of the proposed method: (1) on several synthetic data sets which similarity structures are affected by increasing levels of noise and data sparsity, and (2) on a real data set coming from an acute myeloid leukemia (AML) study. The results obtained are finally compared with the ones of traditional similarity calculation methods. Results In the analysis of the synthetic data set, where the ground truth is known, we measured the capability of reconstructing the correct clusters, while in the AML study we evaluated the Kaplan-Meier curves obtained with the different clusters and measured their statistical difference by means of the log-rank test. In presence of noise and sparse data, our data integration method outperform other techniques, both in the synthetic and in the AML data. Discussion In case of multiple heterogeneous data sources, a matrix trifactorization technique can successfully fuse all the information in a joint model. We demonstrated how this approach can be efficiently applied to discover meaningful patient similarities and therefore may be considered a reliable data driven strategy for the definition of new research hypothesis for precision oncology. Conclusion The better performance of the proposed approach presents an advantage over previous methods to provide accurate patient similarities supporting precision medicine.

Towards a data-integrated cell

“…As building models for each dataset in isolation from others disregards their complementary information, late data integration may result in reduced performance of the integrated model 11,12 . NMTF is an intermediate integration method that directly integrates all datasets through the inference of a single-joint model, which overcomes the above mentioned issues of early and late integration methods, resulting in higher prediction accuracy 12,13 .…”

Fast optimization of non-negative matrix tri-factorization

“…Non-negative matrix tri-factorization is a core component of joint matrix factorization [8] that has been successfully used for fusion of heterogeneous data [55–57]. Such matrix factorization-based data integration can fuse many large datasets [10], however it can require substantial computational resources for inference.…”

Protease target prediction via matrix factorization

“…The algorithm we utilized is a variant of the non-negative matrix trifactorization data integration approach (Vitali et al, 2018(Vitali et al, , 2016 the main difference consisting in cross-validating the training set in order to find the optimal parameter permutation. Whereas in other works factorization ranks have been fixed (Vitali et al, 2018(Vitali et al, , 2016, or selected by optimization, e.g., with the use of a score such as the cophenetic index (Žitnik and Zupan, 2015), in our work we cross-validated sparsenessrelated scaling factors. Note that we also cross-validated the very thresholds utilized for classification.…”

“…repositioning (Vitali et al, 2016), and estimating patient similarity (Vitali et al, 2018). The main advantage of this method, compared to other data integration techniques, is the preservation of the original data structure through the block matrices.…”

“…The main advantage of this method, compared to other data integration techniques, is the preservation of the original data structure through the block matrices. This has been shown to enhance the interpretability of the data integrated (Vitali et al, 2018). On the other hand, the tri-factorization approach requires the definition of initial parameters (e.g., ranks), and might have a high computational demand due to the operations with large, sparse matrices.…”