Neural Genome Sequence Encoding for Downy Mildew Pathogen Detection: A Spatial Pyramid and Dual-Attention Approach

A worldwide crop loss ranging from approximately 6 to 60% in pearl millet has been reported due to downy mildew caused by Sclerospora graminicola. This disease is transmitted through soil or seeds carrying oospores. Field screening, greenhouse experiments, and laboratory techniques are effective in preventing downy mildew in pearl millet; however, these methods are expensive, time-consuming, and require expertise in plant pathology. To address these challenges, researchers have turned to artificial intelligence, employing deep learning techniques such as long-short term memory, and recurrent neural networks, for disease identification using genomic sequences of pathogens. However, existing methods have limitations in interpreting lengthy genomic sequences, preserving the relative positioning of nucleotides, and identifying the minimum sequence length necessary to differentiate one genomic sequence from similar ones. Additionally, these models typically involve numerous training parameters and require significant computation time. Furthermore, none of the existing models specifically focus on recognizing Sclerospora graminicola. In this manuscript, we propose a Neural Genome Sequence Encoder (NeuGSE) model for early prediction of downy mildew in pearl millet using genomic sequences of the fungus Sclerospora graminicola. We also demonstrate the impact of genomic sequence length, positional encoding, and convolutional block attention modules on the performance of the proposed model. Furthermore, we apply two classifiers to prevent halt in training the model due to local minima. Thus, the model achieves the highest precision of 98.04% and an F1 score of 98% on genomic sequence of length 10,000. Moreover, the average F1 score of the proposed model is 19.53% and 13.81% higher than long-short term memory and bidirectional long-short term memory models, respectively, on the same dataset. Additionally, the model can recognize Sclerospora graminicola even in a genomic sequence of length 500. Thus, it may prove valuable in early screening for downy mildew in pearl millet, even before sowing seeds, potentially mitigating crop losses.