Model performance and interpretability of semi-supervised generative adversarial networks to predict oncogenic variants with unlabeled data

Abstract

Background It remains an important challenge to predict the functional consequences or clinical impacts of genetic variants in human diseases, such as cancer. An increasing number of genetic variants in cancer have been discovered and documented in public databases such as COSMIC, but the vast majority of them have no functional or clinical annotations. Some databases, such as CiVIC are available with manual annotation of functional mutations, but the size of the database is small due to the use of human annotation. Since the unlabeled data (millions of variants) typically outnumber labeled data (thousands of variants), computational tools that take advantage of unlabeled data may improve prediction accuracy. Result To leverage unlabeled data to predict functional importance of genetic variants, we introduced a method using semi-supervised generative adversarial networks (SGAN), incorporating features from both labeled and unlabeled data. Our SGAN model incorporated features from clinical guidelines and predictive scores from other computational tools. We also performed comparative analysis to study factors that influence prediction accuracy, such as using different algorithms, types of features, and training sample size, to provide more insights into variant prioritization. We found that SGAN can achieve competitive performances with small labeled training samples by incorporating unlabeled samples, which is a unique advantage compared to traditional machine learning methods. We also found that manually curated samples can achieve a more stable predictive performance than publicly available datasets. Conclusions By incorporating much larger samples of unlabeled data, the SGAN method can improve the ability to detect novel oncogenic variants, compared to other machine-learning algorithms that use only labeled datasets. SGAN can be potentially used to predict the pathogenicity of more complex variants such as structural variants or non-coding variants, with the availability of more training samples and informative features.

BASA-GAN:Self-Attention-Guided Semi-Supervised Balanced Augmented Image Recognition Model

“…For instance, Bu et al [23] developed a semisupervised fine-grained generative adversarial network (SSFGGAN) based on images, sounds, and current to address the issue of evaluating the operational performance of the electro-fused magnesium furnace (EFMF) melting process in the industry. Ren et al [24] introduced a novel semi-supervised generative adversarial neural network (Semi-Supervised Learning) approach to predict carcinogenic variant cells in tasks such as cancer and other diseases. Zhang et al [25] applied DASS to bearing fault diagnosis technology, establishing an intelligent fault diagnosis model for rolling bearings based on an adversarial semi-supervised approach.…”

“…This paper proposes the BASA-GAN model, which takes Semi-supervised GAN [24] as the backbone and integrates autoencoders with the self-attention mechanism. As depicted in figure 3, BASA-GAN comprises two components: the balanced generator module (SAG) and the multi-class discriminator D. Firstly, the labeled real data from the original dataset is converted into random noise and fed into the SAG module.…”

“…We also compare our model with fully supervised models such as CNN, ACGAN, and WGAN. For each original balanced image dataset, we adopt the same settings for the size of the dataset and the division of training and test sets as in [3,19,24]. To illustrate the recognition accuracy of minority classes and demonstrate the generalization performance of the model on imbalanced datasets, we create imbalanced versions of the first two well-known balanced datasets, since SVHN is inherently imbalanced, no modifications are made to it.…”

“…Table 2 shows the overall classification accuracy of the comparative models on datasets A, B, and C. Among them, CNN comprises three convolutional layers and one fully connected layer, while ACGAN [19] and Wasserstein GAN [33] are fully supervised learning models, with experimental settings consistent with the original literature. Self-training [34], DCGAN [3], CCS-GAN [35], scSemiGAN [36], Margin GAN [37], Semi-supervised GAN [24], and the proposed model are semi-supervised learning models. The experimental settings for self-training, DCGAN, CCS-GAN, Margin GAN and scSemiGAN are consistent with the original literature, and the parameter settings for Semi-supervised GAN are the same as the proposed model.…”

Correction: Model performance and interpretability of semi-supervised generative adversarial networks to predict oncogenic variants with unlabeled data

“…Following publication of the original article [ 1 ], it was reported that the article entitled “Model performance and interpretability of semi-supervised generative adversarial networks to predict oncogenic variants with unlabeled data” was published in the regular issue of this journal instead of in the supplement issue.…”