Neural network training with highly incomplete medical datasets

Neural network training and validation rely on the availability of large high-quality datasets. However, in many cases only incomplete datasets are available, particularly in health care applications, where each patient typically undergoes different clinical procedures or can drop out of a study. Since the data to train the neural networks need to be complete, most studies discard the incomplete datapoints, which reduces the size of the training data, or impute the missing features, which can lead to artifacts. Alas, both approaches are inadequate when a large portion of the data is missing. Here, we introduce GapNet, an alternative deep-learning training approach that can use highly incomplete datasets without overfitting or introducing artefacts. First, the dataset is split into subsets of samples containing all values for a certain cluster of features. Then, these subsets are used to train individual neural networks. Finally, this ensemble of neural networks is combined into a single neural network whose training is fine-tuned using all complete datapoints. Using two highly incomplete real-world medical datasets, we show that GapNet improves the identification of patients with underlying Alzheimer’s disease pathology and of patients at risk of hospitalization due to Covid-19. Compared to commonly used imputation methods, this improvement suggests that GapNet can become a general tool to handle incomplete medical datasets.

Omicron-specific mRNA vaccine induced potent neutralizing antibody against Omicron but not other SARS-CoV-2 variants (preprint)

The emerging SARS-CoV-2 variants of concern (VOC) harbor mutations associated with increasing transmission and immune escape, hence undermine the effectiveness of current COVID-19 vaccines. In late November of 2021, the Omicron (B.1.1.529) variant was identified in South Africa and rapidly spread across the globe. It was shown to exhibit significant resistance to neutralization by serum not only from convalescent patients, but also from individuals recieving currently used COVID-19 vaccines with multiple booster shots. Therefore, there is an urgent need to develop next generation vaccines against VOCs like Omicron. In this study, we develop a panel of mRNA-LNP-based vaccines using the receptor binding domain (RBD) of Omicron and Delta variants, which are dominant in the current wave of COVID-19. In addition to the Omicron- and Delta-specific vaccines, the panel also includes a Hybrid vaccine that uses the RBD containing all 16 point-mutations shown in Omicron and Delta RBD, as well as a bivalent vaccine composed of both Omicron and Delta RBD-LNP in half dose. Interestingly, both Omicron-specific and Hybrid RBD-LNP elicited extremely high titer of neutralizing antibody against Omicron itself, but few to none neutralizing antibody against other SARS-CoV-2 variants. The bivalent RBD-LNP, on the other hand, generated antibody with broadly neutralizing activity against the wild-type virus and all variants. Surprisingly, similar cross-protection was also shown by the Delta-specifc RBD-LNP. Taken together, our data demonstrated that Omicron-specific mRNA vaccine can induce potent neutralizing antibody response against Omicron, but the inclusion of epitopes from other variants may be required for eliciting cross-protection. This study would lay a foundation for rational development of the next generation vaccines against SARS-CoV-2 VOCs.

Neural Network Training with Highly Incomplete Datasets (preprint)

Neural network training and validation rely on the availability of large high-quality datasets. However, in many cases only incomplete datasets are available, particularly in health care applications, where each patient typically undergoes different clinical procedures or can drop out of a study. Since the data to train the neural networks need to be complete, most studies discard the incomplete datapoints, which reduces the size of the training data, or impute the missing features, which can lead to artefacts. Alas, both approaches are inadequate when a large portion of the data is missing. Here, we introduce GapNet, an alternative deep-learning training approach that can use highly incomplete datasets. First, the dataset is split into subsets of samples containing all values for a certain cluster of features. Then, these subsets are used to train individual neural networks. Finally, this ensemble of neural networks is combined into a single neural network whose training is fine-tuned using all complete datapoints. Using two highly incomplete real-world medical datasets, we show that GapNet improves the identification of patients with underlying Alzheimer's disease pathology and of patients at risk of hospitalization due to Covid-19. By distilling the information available in incomplete datasets without having to reduce their size or to impute missing values, GapNet will permit to extract valuable information from a wide range of datasets, benefiting diverse fields from medicine to engineering.