PlantDRP predicts disease resistance proteins in plants from protein sequences alone, making large-scale screening accessible without experimental validation.
Plant diseases are responsible for significant yield losses in agricultural crops worldwide, threatening food security for a growing global population. At the heart of plant immunity are disease resistance (R) proteins : a diverse class of immune receptors that enable plants to detect and respond to pathogens including bacteria, fungi, viruses, oomycetes, and nematodes.
R proteins operate through mechanisms such as effector-triggered immunity (ETI) and the hypersensitive response (HR), rapidly containing pathogen spread. The largest and most well-characterised class are the NLR (Nucleotide-binding Leucine-rich Repeat) proteins, which directly or indirectly recognise pathogen-derived effectors.
Experimental identification of R proteins is resource-intensive and time-consuming. With thousands of plant genomes now being sequenced, there is a pressing need for accurate, scalable computational tools to annotate R proteins directly from sequence data : enabling faster resistance breeding and reduced reliance on chemical pesticides.
Upload a protein FASTA file and select a model to obtain per-sequence disease resistance probability scores.
>AT1G12220.1 Arabidopsis thaliana NLR protein MSDNLKQELKELIEQLKKNPAVVKQFLDDIQKEMKDLEDELEAQMKELKDKIEALRQ... >Os01g0700900.1 Oryza sativa resistance protein MASTQQLLLAAAVVVKKNPAVVKQFLDDIQKEMKDLEDELEAQMKELKDKIEALRQ...
> followed by the amino acid sequence.
Minimum sequence length: 50 amino acids.
| # | Sequence ID | Length (aa) | DR Probability | Prediction |
|---|
Two curated datasets were used to train the SVM models. Both are available for download below.
DS1 was adopted directly from the PlantDRPpred tool and comprised 199 experimentally validated plant R-proteins as positive sequences and 199 non-R-proteins as negative sequences, yielding a balanced dataset of 398 sequences. The dataset was partitioned into training and test sets in an 80:20 ratio, resulting in 160 positive and 160 negative sequences in the training set, and 39 positive and 39 negative in the independent test set.
R-protein sequences for DS2 were retrieved from three curated databases: PRGdb 4.0 (152 sequences, 33 plant species), MusaRgeneDB (2,424 sequences after filtering), and RefPlantNLR (481 experimentally validated NLR immune receptors). After filtering and applying CD-HIT at ≤40% similarity, the final dataset comprised 642 positive and 9,759 negative sequences. Training used 500 positive + 500 negative sequences; the independent test set used 142 positive + 142 negative sequences.
PlantDRP is pip-installable. After installation you get a command-line interface for scripted workflows and a Streamlit web UI that runs in your local browser.
--save-embeddings and --no-embed to avoid redundant computation.PlantDRP was developed at the Division of Statistical Genetics, ICAR-Indian Agricultural Statistics Research Institute (ICAR-IASRI), PUSA, New Delhi, India.
The tool applies ProtT5-XL-U50 protein language model embeddings with a Support Vector Machine classifier for sequence-based prediction of plant disease resistance proteins. It requires no structural information or sequence alignment, making it accessible for large-scale genome annotation.
The associated manuscript has been submitted for publication. The citation will be updated upon acceptance.