Table 2 Summary of features considered in models trained to predict mRNA half-life, with a description of the features considered, feature type (i.e., sequence or biochemical), data source, and number of features in the category. Sequence features were calculated identically for both human and mouse. Biochemical features were calculated only with respect to human data
From: The genetic and biochemical determinants of mRNA degradation rates in mammals
Feature description | Feature type | Data source | # features |
|---|---|---|---|
Basic mRNA features such as length and G/C content of 5′ UTR, ORF, and 3′ UTR; intron length; ORF exon junction density | Sequence-derived | Custom Perl scripts | 8 |
Codon frequencies (excluding stop codons) | Sequence-derived | Custom R scripts | 61 |
1- to 7-mer frequencies in the 5′ UTR | Sequence-derived | Custom R scripts | 21,844 |
1- to 7-mer frequencies in the ORF | Sequence-derived | Custom R scripts | 21,844 |
1- to 7-mer frequencies in the 3′ UTR | Sequence-derived | Custom R scripts | 21,844 |
Target predictions for mammalian microRNAs | Sequence-derived | TargetScan predictions [24] | 319 |
Predicted average binding score of human RBPs (in each of 5′ UTR, ORF, and 3′ UTR) | Sequence-derived | DeepRiPe predictions [68] | 177 |
Predicted average binding score of human and mouse RBPs (in each of 5′ UTR, ORF, and 3′ UTR) | Sequence-derived | SeqWeaver predictions [69] | 780 |
Number of CLIP peaks for various RBPs | Biochemical | ENCORI database [70] | 133 |
Number of eCLIP peaks for various RBPs (K562, HepG2, and adrenal gland) | Biochemical | ENCORE database [71] | 225 |
Number of PAR-CLIP peaks | Biochemical | [72] | 146 |
Number of CLIP peaks for m6A pathway components (a6A, m6Am, YTHDF2, METTL3, METTL14, and WTAP) | Biochemical | 13 | |
RIP-seq of diverse RBPs and translational efficiency | Biochemical | 34 |