An autoencoder-based deep learning method for genotype imputation

Abstract

Genotype imputation has a wide range of applications in genome-wide association study (GWAS), including increasing the statistical power of association tests, discovering trait-associated loci in meta-analyses, and prioritizing causal variants with fine-mapping. In recent years, deep learning (DL) based methods, such as sparse convolutional denoising autoencoder (SCDA), have been developed for genotype imputation. However, it remains a challenging task to optimize the learning process in DL-based methods to achieve high imputation accuracy. To address this challenge, we have developed a convolutional autoencoder (AE) model for genotype imputation and implemented a customized training loop by modifying the training process with a single batch loss rather than the average loss over batches. This modified AE imputation model was evaluated using a yeast dataset, the human leukocyte antigen (HLA) data from the 1,000 Genomes Project (1KGP), and our in-house genotype data from the Louisiana Osteoporosis Study (LOS). Our modified AE imputation model has achieved comparable or better performance than the existing SCDA model in terms of evaluation metrics such as the concordance rate (CR), the Hellinger score, the scaled Euclidean norm (SEN) score, and the imputation quality score (IQS) in all three datasets. Taking the imputation results from the HLA data as an example, the AE model achieved an average CR of 0.9468 and 0.9459, Hellinger score of 0.9765 and 0.9518, SEN score of 0.9977 and 0.9953, and IQS of 0.9515 and 0.9044 at missing ratios of 10% and 20%, respectively. As for the results of LOS data, it achieved an average CR of 0.9005, Hellinger score of 0.9384, SEN score of 0.9940, and IQS of 0.8681 at the missing ratio of 20%. In summary, our proposed method for genotype imputation has a great potential to increase the statistical power of GWAS and improve downstream post-GWAS analyses.

Split-Transformer Impute (STI): A Transformer Framework for Genotype Imputation

“…We compare STI to state-of-the-art imputation models: SCDA [43], AE [2], HLA*DEEP [44], and Minimac4 [36]. Additionally, in order to asses the contribution of Cat-Embedding, we replaced it with a convolution layer in STI, named the resulting model STI*WE, fine-tuned it, and applied it to the benchmark datasets.…”

“…In order to benchmark our model, we selected four genotype imputation models: reference-based Minimac4 [36] and reference-free deep learning models SCDA [43], AE [2], and DEEP*HLA [44]. In [43], experimental results indicate superior performance of SCDA to ML models for genotype imputation and as such, we do not repeat the same in this study.…”

“…In [2, 43] studies, the training data is filled with missing values using different percentages, e.g., 0.1, 0.2, etc,. In our experiments, we observed that when 50 percent of the SNPs/SNVs in the training data are randomly replaced with missing values in each iteration, the overall performance of the model is improved, in addition to saving time in the training process.…”

“…Therefore, it is of great importance to limit the scope of attention to save computational resources. Additionally, genotype imputation methods, to the best of our knowledge, are either designed to solely tackle bi-allelic events [36, 41, 42] or are not evaluated for multi-allelic variants [2, 43, 44]. This, in turn, prevents confident imputation of multi-allelic events or complex genetic variants such copy number variants, duplications, and insertions.…”

ProxyTyper: Generation of Proxy Panels for Privacy-aware Outsourcing of Genotype Imputation

“…Overall, our results show that proxy panel generation mechanisms can be used for protecting panels in a task specific manner against honest-but-curious entities and accidental leakages [80], which are most prevalent in terms of the adversarial model in bioinformatics. The mechanisms are flexible and can be integrated into existing pipelines for outsourcing imputation methods, can be used in AI-driven imputation methods [81,82], and in meta-imputation [83] workflows. Moreover, while we focus on imputation as our specific focus, presented mechanisms and new mechanisms can be used for other tasks such as ancestry mapping and kinship estimation.…”