Few-Shot Learning for Predicting Genetic Biomarkers in Cerebral Autosomal Dominant Arteriopathy with Subcortical Infarcts and Leukoencephalopathy (CADASIL)

Laboratory of Integrative Biology, Faculty of Science Ain Chock, Casablanca, University Hassan II, Morocco

^* Corresponding author: This email address is being protected from spambots. You need JavaScript enabled to view it.

Abstract

Cerebral Autosomal Dominant Arteriopathy with Subcortical Infarcts and Leukoencephalopathy (CADASIL) is a rare hereditary cerebral small-vessel disorder with an estimated prevalence of 4.6 per 100,000 adults, primarily caused by NOTCH3 mutations. Its rarity has limited the availability of genetic data, which is important to understand this pathology. Traditional prediction methods require large datasets and fail with limited data. Given these challenges, our study aims to enrich the genetic data on CADASIL. To achieve this, we applied a Few-Shot Learning (FSL) strategy. A total of 4 previously validated CADASIL single nucleotide polymorphisms (SNPs) and 938,544 negative SNPs were extracted from the GWAS catalogue, with their genetic annotations. Based on the assumption of genetic proximity, we generated for each SNP a genomic context string. These strings were embedded into dense vector representations using paraphrase-MiniLM-L6-v2. Similarity scores then ranked candidate SNPs, and the top 100 were identified as novel biomarkers. This in silico framework predicted 100 SNPs and 24 genes. It provides potential biomarkers for early diagnosis, insights into disease mechanisms, and candidate therapeutic targets. This study also validates the compatibility of FSL in the context of rare diseases, paving the way for other applications.