LPI-deepGBDT: a multiple-layer deep framework based on gradient boosting decision trees for lncRNA–protein interaction identification

Abstract Background Long noncoding RNAs (lncRNAs) play important roles in various biological and pathological processes. Discovery of lncRNA–protein interactions (LPIs) contributes to understand the functions mechanisms lncRNAs. Although wet experiments find a few between lncRNAs proteins, experimental techniques are costly time-consuming. Therefore, computational methods increasingly exploited uncover possible associations. However, existing have several limitations. First, majority them were measured based on one simple dataset, which may result prediction bias. Second, applied identify relevant data for new (or proteins). Finally, they failed utilize diverse information proteins. Results Under feed-forward deep architecture gradient boosting decision trees (LPI-deepGBDT), this work focuses classify unobserved LPIs. three human LPI datasets two plant arranged. features proteins extracted by Pyfeat BioProt, respectively. Thirdly, dimensionally reduced concatenated as vector represent an pair. composed forward mappings inverse is developed predict underlying linkages LPI-deepGBDT compared with five classical models (LPI-BLS, LPI-CatBoost, PLIPCOM, LPI-SKF, LPI-HNM) under cross validations lncRNAs, pairs, It obtains best average AUC AUPR values situations, significantly outperforming other identification methods. That is, AUCs computed 0.8321, 0.6815, 0.9073, respectively AUPRs 0.8095, 0.6771, 0.8849, The results demonstrate powerful classification ability LPI-deepGBDT. Case study analyses show that there be GAS5 Q15717, RAB30-AS1 O00425, LINC-01572 P35637. Conclusions Integrating ensemble learning hierarchical distributed representations building multiple-layered architecture, improves performance well effectively probes interaction lncRNAs/proteins.