DeepConv-DTI: Prediction of drug-target interactions via deep learning with convolution on protein sequences
Problem
Identifying drug-target interactions experimentally is costly and labor-intensive, and conventional protein descriptors were insufficiently informative for accurate in silico prediction.
Why it matters
Experimental DTI identification is expensive and slow, hindering drug discovery.
What the AI did
DeepConv-DTI applies a CNN to raw protein sequences, convolving over multiple amino-acid subsequence lengths to capture local residue patterns, trained on large-scale DTI data and evaluated on an independent dataset.
Results
Outperforms previous protein descriptor-based models on an independent dataset; no numeric metrics in abstract
Impact
Offers improved in silico DTI prediction over descriptor-based models, addressing the high cost of experiments.
Record
| Venue | PLoS Computational Biology |
|---|---|
| Published | 2019 |
| Work type | article |
| Authors | Ingoo Lee, Jongsoo Keum, Hojung Nam |
| Citations | 674 (OpenAlex, at import time) |
| References | 62 |
| Model or method | DeepConv-DTI (convolutional neural network) |
| Organization | not stated in source |
| Primary source | https://doi.org/10.1371/journal.pcbi.1007129 |
| Open access | gold |
| Keywords | drug-target interaction, deep learning, CNN, protein sequences, drug discovery |
Importance rating
real but narrower advance
Reduces cost of experimental DTI identification; abstract notes high cost and labor but no scale of savings or affected population.