Network inference in sparse datasets

Abstract

This research addresses the challenge of network inference in large-scale, sparse datasets. Inspired by the Gini coefficient, we introduce a novel methodology leveraging machine learning, including the GENIE3 algorithm and Random Forest, for efficient network inference. Principal Component Analysis (PCA) is employed for dimension reduction, with systematic evaluations across varied sparseness levels. Notably, our study incorporates an in-house developed backprojecting methodology, enhancing the preservation of feature importances during the inference process. The adapted Gini-based algorithm, coupled with the proprietary backprojecting methodology, showcases promising results in sparse network inference and comparative analyses with SVD and NMF. This research contributes to the acceleration and reliability of network inference, offering a valuable advancement in the field.