MODEM is a comprehensive database of maize multidimensional omics data, including genomic, transcriptomic, metabolic and phenotypic information from the cellular to individual plant level. This initial release contains approximately 1.06 M high quality SNPs for 508 diverse inbred lines obtained by combining variations from RNA sequencing on whole kernels (15 days after pollination) of 368 lines and a 50 K array for all 508 individuals. As all of these data were derived from the same diverse panel of lines, the database also allows various types of genetic mapping (including characterization of phenotypic QTLs, pQTLs; expression QTLs, eQTLs and metabolic QTLs, mQTLs). MODEM is thus designed to promote a better understanding of maize genetic architecture and deep functional annotation of the complex maize genome (and potentially those of other crop plants) and to explore the genotype–phenotype relationships and regulation of maize kernel development at multiple scales, which is also comprehensive for developing novel methods. MODEM is additionally designed to link with other databases to make full use of current resources, and it provides visualization tools for easy browsing. All of the original data and the related mapping results are freely available for easy query and download. This platform also provides helpful tools for general analyses and will be continually updated with additional materials, features and public data related to maize genetics or regulation as they become available. Database URL: ( http://modem.hzau.edu.cn )
The integration of genotypic and environmental data can enhance the prediction accuracy of field traits of crops. The existing genomic prediction methods fail to consider the environmental factors and do not consider the real growing environment of crops, resulting in low genomic prediction accuracy. In this work, we propose a genotype-environment interaction genomic prediction method in maize, called GEFormer, based on integrating the gating mechanism MLP and linear attention mechanism. Firstly, it uses gated multilayer perceptron (gMLP) to extract the local and global features among SNPs. Then, the Omni-dimensional Dynamic Convolution is used to extract the dynamic and comprehensive features of multiple environmental factors within each day in the consideration of the real growth pattern of crops. The linear attention mechanism is used to capture the temporal features of environmental changes. Finally, it uses the gating mechanism to fuse the genomic and environmental features effectively. We validate the accuracy of GEFormer in predicting important agronomic traits of maize, rice and wheat in three experimental scenarios: untested genotypes in tested environments, tested genotypes in untested environments, untested genotypes in untested environments. Experimental results show that GEFormer outperforms six cutting-edge statistical learning methods and four machine learning methods. Furthermore, it shows great advantages in the experimental scenario of untested genotypes in untested environments. In addition, we used GEFormer into three real-world breeding applications: phenotype prediction in unknown environments, hybrid phenotype prediction using inbred population, and cross-population phenotype prediction. The results illustrate that GEFormer exhibiting better prediction performance in actual breeding scenarios, and it can be utilized to assist crop breeding.
Mining epistatic loci which affects specific phenotypic traits is an important research issue in the field of biology. Bayesian network (BN) is a graphical model which can express the relationship between genetic loci and phenotype. Until now, it has been widely used into epistasis mining in many research work. However, this method has two disadvantages: low learning efficiency and easy to fall into local optimum. Genetic algorithm has the excellence of rapid global search and avoiding falling into local optimum. It is scalable and easy to integrate with other algorithms. This work proposes an epistasis mining approach based on genetic tabu algorithm and Bayesian network (Epi-GTBN). It uses genetic algorithm into the heuristic search strategy of Bayesian network. The individual structure can be evolved through the genetic operations of selection, crossover and mutation. It can help to find the optimal network structure, and then further to mine the epistasis loci effectively. In order to enhance the diversity of the population and obtain a more effective global optimal solution, we use the tabu search strategy into the operations of crossover and mutation in genetic algorithm. It can help to accelerate the convergence of the algorithm. We compared Epi-GTBN with other recent algorithms using both simulated and real datasets. The experimental results demonstrate that our method has much better epistasis detection accuracy in the case of not affecting the efficiency for different datasets. The presented methodology (Epi-GTBN) is an effective method for epistasis detection, and it can be seen as an interesting addition to the arsenal used in complex traits analyses.
Residual heterozygosity (RH) in advanced inbred lines of plants benefits quantitative trait locus (QTL) mapping studies. However, knowledge of factors affecting the genome-wide distribution of RH remains limited. A set of 2196 heterogeneous inbred family (HIF) maize lines derived from 12 recombinant inbred line (RIL) populations was genotyped using the Maize50K SNP chip. A total of 18,615 unique RH intervals were identified, ranging from 505 to 2095 intervals per population, with average maize genome coverage of 94.8%. Across all populations, there were 8.6 RH intervals per HIF line on average, ranging from 1.8 to 14 intervals; the average size of an RH interval was approximately 58.7 Mb, ranging from 7.2 to 74.1 Mb. A given RH region was present in an average of 5 different individuals within a population. Seven RH hotspots, where RH segments were enriched in the genome, were found to be subject to selection during population development. The RH patterns varied significantly across populations, presumably reflecting differences in the genetic background of each population, and 8 QTLs were found to affect heterozygosity levels in the RH hotspots. The potential use of this HIF library for the fine mapping of QTLs was assessed based on publicly available QTL information, achieving a ≤ 1 Mb resolution on average. The examined library of HIF lines offers insight into the RH landscape and its intraspecific variation and provides a useful resource for the QTL cloning of important agronomic traits in maize.
Identifying the function of DNA sequences accurately is an essential and challenging task in the genomic field. Until now, deep learning has been widely used in the functional analysis of DNA sequences, including DeepSEA, DanQ, DeepATT and TBiNet. However, these methods have the problems of high computational complexity and not fully considering the distant interactions among chromatin features, thus affecting the prediction accuracy. In this work, we propose a hybrid deep neural network model, called DeepFormer, based on convolutional neural network (CNN) and flow-attention mechanism for DNA sequence function prediction. In DeepFormer, the CNN is used to capture the local features of DNA sequences as well as important motifs. Based on the conservation law of flow network, the flow-attention mechanism can capture more distal interactions among sequence features with linear time complexity. We compare DeepFormer with the above four kinds of classical methods using the commonly used dataset of 919 chromatin features of nearly 4.9 million noncoding DNA sequences. Experimental results show that DeepFormer significantly outperforms four kinds of methods, with an average recall rate at least 7.058% higher than other methods. Furthermore, we confirmed the effectiveness of DeepFormer in capturing functional variation using Alzheimer’s disease, pathogenic mutations in alpha-thalassemia and modification in CCCTC-binding factor (CTCF) activity. We further predicted the maize chromatin accessibility of five tissues and validated the generalization of DeepFormer. The average recall rate of DeepFormer exceeds the classical methods by at least 1.54%, demonstrating strong robustness.