Our research efforts are focused on developing novel computational algorithms, computational models, and computational pipelines for big, heterogeneous biomedical and clinical data. Here, we briefly present some examples. The DeepEar Project The detection of congenital auricular deformity within 6 months after birth is necessary to avoid later surgical interventions. Currently, there are no reliable objective methods to detect ear deformity; subjective assessment is the only way. Therefore, we developed a deep learning model using a transfer learning technique to classify ears from 2D photographs as normal or deformed with high accuracy. The methodology is depicted in the graphic below from our manuscript Identifying ear abnormality from 2D photographs using convolutional neural networks published in Scientific Reports in 2019 (authors Rami R. Hallac, Jeon Lee, Mark Pressler, James R. Seaward, and Alex A. Kane). ECMO Injury Prediction Project Brain injury is a significant source of morbidity and mortality for pediatric patients treated with Extracorporeal Membrane Oxygenation (ECMO). To predict brain injury in pediatric ECMO-supported patients, we developed a hybrid deep learning model consisting of a recurrent neural network (LSTM) and a convolutional neural network (CNN). Taking time series data comprised of physiological features, markers of end-organ perfusion, acid-base homeostasis, vasoactive infusions, markers of coagulation, and ECMO-machine factors, our deep learning model demonstrated accurate predictions that surpassed clinician judgements regarding pediatric patients on ECMO that were more likely to develop brain injury. Details of this study can be found within our manuscript Neural networks to predict radiographic brain injury in pediatric patients treated with extracorporeal membrane oxygenation published in 2020 in the Journal of Clinical Medicine (authors Neel Shah, Abdelaziz Farhat, Jefferson Tweed, Ziheng Wang, Jeon Lee, Rafe McBeth, Michael Skinner, Fenghua Tian, Ravi Thiagarajan, and Lakshmi Raman). HIV-1 Proviral Expression Project The impact of HIV-1 integration site on proviral transcription and expression is not well understood and requires the analysis of multiple genomic datasets for thousands of proviral integration sites across the human genome. We generated and combined large-scale datasets, including epigenetics, transcriptome, and 3D genome architecture to interrogate the chromatin states, transcription activity, and nuclear sub-compartments around HIV-1 integrations in Jurkat CD4+ T-cells to decipher human genome regulatory features shaping proviral transcription. Using machine learning approaches, we defined the nuclear sub-compartments and chromatin states that shape genomic architecture, transcriptional activity, and nucleosome density of neighboring regions around the integration site. These were all additive features influencing HIV-1 expression. Based on this integrative genomics approach, we can now introduce HIV-1 at targeted genomic locations having known regulatory features and study how such integrations influence viral expression. More details in our manuscript Decoding Human Genome Regulatory Features That Influence HIV-1 Proviral Expression and Fate Through an Integrated Genomics Approach published in Bioinformatics and Biology Insights in 2022 (authors Holly Ruess, Jeon Lee, Carlos Guzman, Venkat S. Malladi, and Iván D'Orso). Human Prostate Anatomy Project The cellular origins of benign prostatic hyperplasia (BPH) and prostate cancer are still unknown. To properly define human prostate cellular anatomy and create a baseline for understanding the cellular origins of disease, we performed single-cell RNA sequencing (scRNA-seq) on single cells from five young adult human prostates. After multiple scRNA-seq data normalization and integration, two unrecognized epithelial cell types were newly identified, and previously unknown markers were derived for established cell types. Details can be found in our manuscript A cellular anatomy of the normal adult human prostate and prostatic urethra published in Cell Reports in 2018 (authors Gervaise H. Henry, Alicia Malewska, Diya B. Joseph, Venkat S. Malladi, Jeon Lee, Jose Torrealba, Ryan J. Mauck, Jeffrey C. Gahan, Ganesh V Raj, Claus G. Roehrborn, Gary C. Hon, Malcolm P. MacConmara, Jeffrey C. Reese, Ryan C. Hutchinson, Chad M. Vezina, and Douglas W. Strand). Prediction of Mechanism of Action Project Assigning the accurate mechanism of action to botanicals, natural products, and synthetic chemicals continues to be a major challenge. We integrated gene expression-based (FUSION) and high content imaging-based (Cytological Profiling, CP) screening platforms using Similarity Network Fusion (SNF) that resulted in a novel, improved framework for functional annotation of natural products. We validated the utility of this computational prediction by confirming and demonstrating that natural product factions containing Trichostatin A were clustered with pure Trichostatin A and other HDAC inhibitors. This work is described in our manuscript High-throughput functional annotation of natural products by integrated activity profiling on bioRxiv in 2019 (authors Suzie K. Hight, Kenji L. Kurita, Elizabeth A. McMillan, Walter Bray, Trevor N. Clark, Anam F. Shaikh, F. P. Jake Haeckl, Fausto Carnevale-Neto, Scott La, Akshar Lohith, Rachel M. Vaden, Jeon Lee, Shuguang Wei, R. Scott Lokey, Michael A. White, Roger G. Linington, and John B. MacMillan). Prediction of Treatment Outcome after Multi-Modality Fusion For patients with Major Depressive Disorder (MDD), predicting a patient's drug response is crucial to guide drug selection and improve clinical outcomes. The existing drug selection process is suboptimal because it is based on trial-and-errors, each of which usually takes 3 months to observe an effect. In this project, we are developing an end-to-end deep learning framework to predict anti-depressant treatment outcomes by integrating pre-treatment brain activity and connectivity measured from EEG and fMRI. This is an ongoing collaboration project with the Deep Learning for Precision Health Lab (PI: Albert Montillo).