zhuangyh

AML_classifier

As a health data science fellow at Insight, Boston, I have developed an automatic diagnosis tool for of Acute Myeloid Leukemia (one of blood cancer) using flow cytometry Data. The accuracy of predicted diagnosis is about 94%.

Dysregulation_of_gene_expressions_upon_Zika_virus_infection

Identified cellular genes and signaling pathways that are dysregulated in the mouse brain following infection with ZIKV with Next generation RNA sequencing data.

Augmented_Bayesian_eQTL_algorithms

A novel tissue augmented Bayesian model for eQTL analysis (TA-eQTL) was developed to integrate prior eQTL information from one tissue and better predict eQTL within a tissue.

SAS-Demographic-Table

SAS codes to produce Table, Listing, and Figure in a clinical trial setting.

Longitudinal-Data-Analysis

The data for this report are from the Diabetes Control and Complications Trial (DCCT) conducted between 1983 and 1986. Patients with type 1 diabetes (T1D) were recruited from 29 medical centers in the U.S. and Canada, were randomized to intensive or standard therapy, and asked to complete physical exam visits for 9 years. In this study, the objective was to identify the trajectories of weight gain, which has not been reported yet. The investigators conceptualized that the first 3 years represents acute period in weight gain and later years represent a maintenance period. The primary questions of interest were whether the rates of change in weight gain differ for the treatment groups, whether the rate of weight gain differs before and after 3 years of follow-up, and how A1C influences changes in weight over time. To access the above questions, we hypothesized 1) that the association between time and rates of change in weight gain significantly depends on treatments and the annual rates of weight gain differ between the first 3 year of treatment (acute period) and after 3 years follow-up (maintenance period) in both two treatment groups; 2) that HBA1C has mediation effects and the difference of annual change between two treatment groups in acute period and/or maintenance period would shrink after adjusting HBA1C. To test the above hypothesis, the primary outcome would be BMI and the predictors include time, treatments and other necessary factors. In addition, it is known that the investigators picked 3 years only because it seemed like most of the weight gain happened in the first three years. Thus, the secondary question of interest in this study was whether a change at 3 years was the right cut point.

SAS-Disposition-Table

Survival_Analysis

The incidence of malignant melanoma has been on the rise at an alarming rate in the United States and attracted great attention (1). The occurrence of melanoma is associated with race, skin color, skin tendency to burn, freckles, blue or green eye color, light hair color, family history, and prevalence of numerous melanocytic nevi (2-4). It has been shown that the presence of numerous melanocytic nevi is the strongest risk factor for melanoma. Nevi are likely to be precursor lesions for 20-60% of melanomas (5). White populations have higher risks for malignant melanoma than other racial/ethnic groups. In the United States, non-Hispanic white individuals had an annual incidence rate of 25.1 per 100 000 population for the period 2000 through 2004 compared with 1.0 per 100 000 for black, 4.5 per 100 000 for Hispanic white (1, 6). Two studies suggested heritability accounts for about 2/3 of the variance in nevus counts (7) (8). Several specific genetic variations have been implicated. The most notable genetic factors implicated in melanoma at present include CDKN2A, MC1R, and OCA2 (9). It has been shown that the OCA2 rs12913832 SNP is associated with Caucasian populations (10). OCA2 was also strongly related with hair color, with 36% of those homozygous for the g form (gg) having blonde hair compared to 8% of homozygotes for the  form (aa) (10). Given the strong relationship between nevus density and melanoma risk, melanoma risk genes are likely candidate genes for nevus formation. Thus, we hypothesized that OCA2 has association with nevus and that population with certain OCA2 variant are susceptible to developing nevus. To determine the influence of OCA2 on total nevus counts and changes on children, we also take gender and race into account in this study during estimating the effect of OCA2 on nevus development.

SAS-Adverse-Events-Table

SAS codes to generate two adverse event(AE) tables

easyVAF

Somatic sequence variants are associated with a cancer diagnosis, prognostic stratification, and treatment response. Variant allele frequency (VAF) is the percentage of sequence reads with a specific DNA variant over the read depth at that locus. VAFs on targeted loci under different (experimental) conditions are often compared. We present our R package ‘ esayVAF’ for parametric and non-parametric comparison of VAFs among multiple treatment groups.

SAS-Lab-Shift-Table

zhuangyh/AhGlasso

The repository includes the key R source codes for AhGlasso algorithm.

zhuangyh/twoOmicGNN

This repository includes source codes of two-omics GNN and a Jupyter notebook for the paper entitled "Deep learning on graphs for multi-omics classification of COPD".

zhuangyh/easyVAF

Somatic sequence variants are associated with a cancer diagnosis, prognostic stratification, and treatment response. Variant allele frequency (VAF) is the percentage of sequence reads with a specific DNA variant over the read depth at that locus. VAFs on targeted loci under different (experimental) conditions are often compared. We present our R package ‘ esayVAF’ for parametric and non-parametric comparison of VAFs among multiple treatment groups.

zhuangyh/gym

A toolkit for developing and comparing reinforcement learning algorithms.

zhuangyh/deep-reinforcement-learning

Repo for the Deep Reinforcement Learning Nanodegree program

zhuangyh/Flower-Image-Classifier-Project-Pytorch-Challenge

This project is for Facebook/Udacity Pytorch challenge. I built an image classifier based on pretrained model (desnet121) to identify different species (102) of flowers. With 18 epochs training, 95.5% accuracy in the validation data set was achieved.

zhuangyh/SmCCNet_KK

A canonical correlation analysis based method for discovering (quantitative) trait-specific heterogeneous regulatory networks.

zhuangyh/Bayesian-inference

zhuangyh/DeepLearning.ai-Summary

This repository contains my personal notes and summaries on DeepLearning.ai specialization courses. I've enjoyed every little bit of the course hope you enjoy my notes too.

zhuangyh/AML_classifier

As a health data science fellow at Insight, Boston, I have developed an automatic diagnosis tool for of Acute Myeloid Leukemia (one of blood cancer) using flow cytometry Data. The accuracy of predicted diagnosis is about 94%.

zhuangyh/Augmented_Bayesian_eQTL_algorithms

A novel tissue augmented Bayesian model for eQTL analysis (TA-eQTL) was developed to integrate prior eQTL information from one tissue and better predict eQTL within a tissue.

zhuangyh/Dysregulation_of_gene_expressions_upon_Zika_virus_infection

Identified cellular genes and signaling pathways that are dysregulated in the mouse brain following infection with ZIKV with Next generation RNA sequencing data.

zhuangyh/Survival_Analysis

The incidence of malignant melanoma has been on the rise at an alarming rate in the United States and attracted great attention (1). The occurrence of melanoma is associated with race, skin color, skin tendency to burn, freckles, blue or green eye color, light hair color, family history, and prevalence of numerous melanocytic nevi (2-4). It has been shown that the presence of numerous melanocytic nevi is the strongest risk factor for melanoma. Nevi are likely to be precursor lesions for 20-60% of melanomas (5). White populations have higher risks for malignant melanoma than other racial/ethnic groups. In the United States, non-Hispanic white individuals had an annual incidence rate of 25.1 per 100 000 population for the period 2000 through 2004 compared with 1.0 per 100 000 for black, 4.5 per 100 000 for Hispanic white (1, 6). Two studies suggested heritability accounts for about 2/3 of the variance in nevus counts (7) (8). Several specific genetic variations have been implicated. The most notable genetic factors implicated in melanoma at present include CDKN2A, MC1R, and OCA2 (9). It has been shown that the OCA2 rs12913832 SNP is associated with Caucasian populations (10). OCA2 was also strongly related with hair color, with 36% of those homozygous for the g form (gg) having blonde hair compared to 8% of homozygotes for the  form (aa) (10). Given the strong relationship between nevus density and melanoma risk, melanoma risk genes are likely candidate genes for nevus formation. Thus, we hypothesized that OCA2 has association with nevus and that population with certain OCA2 variant are susceptible to developing nevus. To determine the influence of OCA2 on total nevus counts and changes on children, we also take gender and race into account in this study during estimating the effect of OCA2 on nevus development.

zhuangyh/Longitudinal-Data-Analysis

The data for this report are from the Diabetes Control and Complications Trial (DCCT) conducted between 1983 and 1986. Patients with type 1 diabetes (T1D) were recruited from 29 medical centers in the U.S. and Canada, were randomized to intensive or standard therapy, and asked to complete physical exam visits for 9 years. In this study, the objective was to identify the trajectories of weight gain, which has not been reported yet. The investigators conceptualized that the first 3 years represents acute period in weight gain and later years represent a maintenance period. The primary questions of interest were whether the rates of change in weight gain differ for the treatment groups, whether the rate of weight gain differs before and after 3 years of follow-up, and how A1C influences changes in weight over time. To access the above questions, we hypothesized 1) that the association between time and rates of change in weight gain significantly depends on treatments and the annual rates of weight gain differ between the first 3 year of treatment (acute period) and after 3 years follow-up (maintenance period) in both two treatment groups; 2) that HBA1C has mediation effects and the difference of annual change between two treatment groups in acute period and/or maintenance period would shrink after adjusting HBA1C. To test the above hypothesis, the primary outcome would be BMI and the predictors include time, treatments and other necessary factors. In addition, it is known that the investigators picked 3 years only because it seemed like most of the weight gain happened in the first three years. Thus, the secondary question of interest in this study was whether a change at 3 years was the right cut point.

zhuangyh/SAS-Demographic-Table

SAS codes to produce Table, Listing, and Figure in a clinical trial setting.

zhuangyh/SAS-Lab-Shift-Table

zhuangyh/SAS-Adverse-Events-Table

SAS codes to generate two adverse event(AE) tables

zhuangyh/SAS-Disposition-Table