<reflect on unemployment if possible- for example, indicate that the reason to go for msc and/or phd is really unemployment>
<reflect on the demographies in relation to dropout/ graduation>
<reflect on logistics in the results/ discussion section>
<reflect on logistics in the results/ discussion section>
<reflect on training of local TAs later on: positive affirmations and  how they help solve problems>
<to drive this point home, you need to show correlation between successful IBT graduates and their career level: at least whether current students were more likely to graduate or not. look for gender or career bias>
<Is there correlation between how people heard of the course and either their university or affiliation..>
<reflect on challenges relating to setting up the place, running the course>
<reflect on the waiting list from run to run of the IBT>
Consistent with the largely female students' body composition in faculties related to natural sciences, health, and agriculture in Sudan \cite{huyer2015gender}, faculties from which most of the 2017 IBT participants hail, the majority of the IBT participants in both classes of the H3ABioNet node of Sudan were females (80%)and they were also more likely to satisfy the course requirements in comparison with their male peers  (Supplementary_Figure3_Class_demographics_and_performance_distribution). This remark, about gender as a strong demographic predictor of an IBT participant performance in the course can also be seen from the rpart recursive partitioning tree \cite{rpart} of (Supplementary_Figure4_rpart_performance_classification_tree & Supplementary_Text1_rpart_tree_specifications), built by selecting spiting covariates that minimize the Gini coefficient as an information measure.
However, when changing the partitioning algorithm to a conditional inference framework, as implemented in the party package in R  \cite{hothorn2006unbiased},  it is noted that the location of the local IBT classroom (the CBSB lab or the Main Library), is the most important covariate in predicting the performance of participants based on the complete dataset (Supplementary_Figure6_Conditional_Inference_Tree_for_Participants_Performance_in_the_IBT). This is also not surprising considering the inherent differences between the 2 locations in terms of infrastructure (the CBSB lab is designed to facilitate bioinformatics training and research in terms of infrastructure with stable internet connection and more powerful computers; whereas at the beginning of the course, the Main Library classroom had logistic problems in terms of Internet connectivity, which was frustrating to some of the participants (and in occasions encouraging some of them to ultimately withdraw early on)).
The difference in the trees constructed by these algorithms (Supplementary_Figure4_rpart_performance_classification_tree &  Supplementary_Figure6_Conditional_Inference_Tree_for_Participants_Performance_in_the_IBT),  can be explained by the high degree of class imbalance in the entire dataset (70% Success, 26% Withdrawal and 4% Failure) , and  especially when seen in each classroom independently (85%, 15% and  0% for the CBSB class, and 58%, 35% and 8% for the Main Library respectively). We therefore built a multinomial classification model to further examine all demographic factors collectively (Table   ,  Figure ; Supplementary)