Inferring microRNA regulation: A proteome perspective

Abstract

Post-transcriptional regulation in multicellular organisms is mediated by microRNAs. However, the principles that determine if a gene is regulated by miRNAs are poorly understood. Previous works focused mostly on miRNA seed matches and other features of the 3′-UTR of transcripts. These common approaches rely on knowledge of the miRNA families, and computational approaches still yield poor, inconsistent results, with many false positives. In this work, we present a different paradigm for predicting miRNA-regulated genes based on the encoded proteins. In a novel, automated machine learning framework, we use sequence as well as diverse functional annotations to train models on multiple organisms using experimentally validated data. We present insights from tens of millions of features extracted and ranked from different modalities. We show high predictive performance per organism and in generalization across species. We provide a list of novel predictions including Danio rerio (zebrafish) and Arabidopsis thaliana (mouse-ear cress). We compare genomic models and observe that our protein model outperforms, whereas a unified model improves on both. While most membranous and disease related proteins are regulated by miRNAs, the G-protein coupled receptor (GPCR) family is an exception, being mostly unregulated by miRNAs. We further show that the evolutionary conservation among paralogs does not imply any coherence in miRNA regulation. We conclude that duplicated paralogous genes that often changed their function, also diverse in their tendency to be miRNA regulated. We conclude that protein function is informative across species in predicting post-transcriptional miRNA regulation in living cells.

Automated Annotation of Disease Subtypes

“…A concern was whether the model was simply identifying surface-level patterns, such as associating all diseases from a specific data-source, such as Orphanet (an orphan disease database) with a subtype. Such hidden confounders are common in many predictive scenarios [35], [36]. To validate our model, we examined the distribution of disease subtypes in our predictions against the known subtypes, focusing on their source database ( Table 1 ).…”

“…These features are denoted as “Text Embedding X’’ in Fig 3, where X represents a vector in the embedding. For interpretability, we implemented an automated explanation framework showing exemplars of high, neutral and low values for embedding dimensions, inspired by approaches in other NLP or automated-machine learning works[35], [36], [52]. It is available in our codebase.…”

“…For computational reasons, the genetic associations and phenotype sources were filtered to keep only those appearing at least 30 times in the dataset, and these were used as features, including their evidence scores. A novel approach [19], [36] of compressing these sparse features using a truncated singular matrix decomposition representation of ~512 dimensions worked well in terms of performance and compute (not shown), but reduced interpretability, and thus was not used in the final model analysis. "Phenotype counts" is the number of distinct phenotypes associated with a disease (regardless of individual phenotype frequency or evidence score).…”

Protein Language Models Expose Viral Mimicry and Immune Escape

“…Multiple partitions of the data were analyzed separately: the entire dataset (25,117 sequences), a human-only subset (18,418 sequences), and only viruses from genera with a human host (3,915 sequences). Features were extracted using the SparkBeyond autoML framework, as described in (Ofer and Linial, 2022). Inputs included the protein sequence, length, taxonomy, name, UniProt keywords, virus-host species, and Baltimore classification, but not embeddings.…”

Automated Street Light Adjustment System on Campus with AI-Assisted Data Analytics

“…Functionalities in the SparkBeyond Discovery platform (SparkBeyond, Israel) were adopted to provide an AI-driven engine as a tool for extraction of meaningful features and to perform experiments on various machine learning models to learn and solve the time-series problem. The tool has been successfully applied in many research works to gain insight and create efficient models [ 28 , 29 , 30 ], as well as several industrial applications, ranging from banking, and electronic commerce, to insurance.…”