In this study, we propose a new predictor named iPiDA-LTR (see Fig.1) to identify piRNA-disease associations, which has the following advantages:
(i) iPiDA-LTR predictor combines component methods and learning to rank,
which cannot only identify missing associations between known piRNAs and diseases,
but also can identify diseases associated with newly detected piRNAs;
(ii) The task of identifying piRNA-disease associations is a positive unlabelled learning problem.
The iPiDA-LTR predictor incorporates learning to rank only considering the top ranked positive samples;
(iii) A free web server of iPiDA-LTR has been established at http://bliulab.net/iPiDA-LTR.
Fig.1.
The workflow of iPiDA-LTR predictor.
(a) Association feature extraction:
piRNA sequences and disease ontology are used to calculate piRNA sequence similarities and disease semantic similarities by combining them and piRNA-disease associations to construct association features and labels;
(b) Component methods: four methods are used to train models with benchmark dataset, and then trained models are utilized to calculate association scores of query piRNAs;
(c) Predicting candidate diseases associated with query piRNAs:
association scores of samples in the benchmark dataset are used to train LambdaMART model, and then trained LambdaMART model are employed to rank candidate diseases associated with query piRNAs.
