RNAm5CPred

RNAm5CPred predicts RNA 5-methylcytosine (m5C) modification sites using support vector machine (SVM)-based classification of sequence-derived features. It integrates nucleotide composition descriptors to identify m5C sites across RNA sequences.

Key Features:

• Sequence Feature Integration: Utilizes K-nucleotide frequencies (KNFs, including 4NF), K-spaced nucleotide pair frequencies (KSNPFs, including 1SNPF), and pseudo dinucleotide composition (pseDNC) to capture local and global sequence patterns.
• SVM-Based Classification: Implements support vector machine models trained on redundant, nonredundant, balanced (Met240), and imbalanced (Met935) datasets.
• Jackknife Validation: Applies jackknife testing to identify optimal feature combinations, achieving improved performance metrics including recall and Matthews correlation coefficient (MCC).
• Independent Test Evaluation: Demonstrates performance on independent dataset Test1157, with highest recall (68.79%) and MCC (0.154) from the Met240-trained model.

Scientific Applications:

• Epitranscriptomic Site Identification: Enables computational prediction of m5C modification sites to support functional studies of RNA methylation.

Methodology:

RNAm5CPred extracts KNF, KSNPF, and pseDNC features from RNA sequences, trains SVM classifiers on curated datasets with varying redundancy and class balance, evaluates models using jackknife and independent testing, and selects optimal feature combinations for m5C site prediction.