- CNN-LSTM model on DNA sequence
- Hybrid CNN and bidirectional LSTM RNN structure to predict DNA sequences to binary output
- Achieved 94% test accuracy
- Multimodal-MARS: Variational autoencoder based models for single cell annotation
- Incorporated VAE and CVAE models into MARS-single cell autoencoder based model (Brbic et al. 2020)
- Achieved improved modality integration between single cell RNA and ATAC seq data
- Enhanced label transfering from single cell RNA to highly sparsed ATAC seq data
- Achieved better clustering boundaries than AE based MARS for multi-modal single cell annotation
- GRU-based multi head model to predict microbe-host and microbe-microbe interactions
- Achieved multi-tasking classification of host disease state
- Achieved autoregressive time seires prediction of microbial relative abundance at different taxonomy levels
- Created in-silico experiments to test perturbation effects
