TABLE 6.
Description of the train/test datasets, feature encoding and machine learning strategy for each of the described methods.
| Method | Interacting molecules | Train/test dataset | Feature encoding | Machine learning strategy | References |
|---|---|---|---|---|---|
| LPI-deepGBDT | lncRNA-RBP | Derived from NPInter | Sequence features extracted using Pyfeat (Muhammod et al., 2019) and BioProt (Márquez and Castro Amaya, 2019) | Gradient boosting decision trees | Zhou et al. (2021) |
| LncPNet | lncRNA-RBP | Derived from NPInter v2.0 | Heterogeneous network embedding of lncRNAs and proteins similarity networks and of the known lncRNA-protein interaction network | Support-vector machine | Zhao et al. (2021) |
| CRBPDL | circRNA-RBP | CLIP-seq experiments | k-nucleotide frequency (KNF), Doc2vec, electron-ion interaction pseudopotential (EIIP), chemical characteristics of nucleotides (CCN) and accumulated nucleotide frequency (ANF) | Deep multi-scale residual network (ResNet) and bidirectional gated recurrent unit with a self-attention mechanism (BiGRUs) | Niu et al. (2022) |
| EDLMFC | ncRNA-RBP | RPI1807 NPInter v2.0 RPI488 | k-mer frequencies of the sequence and structure representations | Ensemble deep learning framework including convolutional neural networks (CNN) and bi-directional long short-term memory net-work (BLSTM) | Wang et al. (2021) |
| preMLI | miRNA-mRNA | Plants lncRNA-miRNA interaction dataset constructed using RNAHybrid 2.1.2 | word2vec based sequence embedding | CNN and bidirectional gated recurrent unit (Bi-GRU) | Yu et al. (2022) |
| PrismNet | RNA-RBP | CLIP-seq experiments | One-hot-encoded sequence vectors and icSHAPE structure scores | Convolutional layers, squeeze-and-excitation networks (SE) and residual blocks | Sun et al. (2021) |
| PRNA | RNA-RBP | RsiteDB | Number of atoms, electrostatic charge, potential hydrogen bonds, hydrophobicity and relative accessible surface area were used as sequence features. Secondary structure of amino acid residues, conservation score (PSI-BLAST), side-chain environment were used as structure features. A sliding window was used to encode amino acid residues and create feature vectors | Random Forest | Liu et al. (2010) |