2
$\begingroup$

I am a computer science student, who was interested in taking research in the field of bioinformatics.

The first research idea I had was to make a model that could predict the possibility of type 2 diabetes by using machine learning such as convolutional neural networks or recurrent neural networks by training the model with fasta data of protein that causes diabetes.

The research will focus more on the architecture of the machine learning itself.

Does this research possible to work on ? If yes, how and where can I get the data? All my lecturers have no background in bioinformatic, so I can't ask any of them.

Edit: I m sorry, not protein but gene like CAPN10, CDKAL1, CDKN2A, HHEX, HNF4A, IGF2BP2, KCNJ11, PPARG, SLC30A8, and TCF7L2 that shown have risk to be associated in Type 2 Diabetes (http://diabetes.diabetesjournals.org/content/56/12/3033).

$\endgroup$
  • 1
    $\begingroup$ Before you can do anything valuable with machine learning in bioinformatics you must understand the biology. You cannot select meaningful features if you do not know what they mean. There is no protein that causes diabetes, and if there was its fasta sequence would most likely not be a very helpful feature. $\endgroup$ – Wouter De Coster Dec 2 '18 at 21:48
  • $\begingroup$ I m sorry, not protein but gene like CAPN10, CDKAL1, CDKN2A, HHEX, HNF4A, IGF2BP2, KCNJ11, PPARG, SLC30A8, and TCF7L2 that shown have risk to be associated in Type 2 Diabetes (diabetes.diabetesjournals.org/content/56/12/3033). The model will be trained by those gene, both from the infected and healthy sample,it is still possible? Thank you for answering $\endgroup$ – Nurrakman Dwinanda W Dec 2 '18 at 22:21
  • $\begingroup$ Are you familiar with GWAS, linkage disequilibrium, functional vs meaningless variants, sequencing and variant calling? If not, that would be your starting point. If so, you would know that the variants identified are most likely non-coding and not the true functional variant. Follow-up studies, ideally including functional modeling are necessary to figure out those variants which actually affect diabetes and which are just on the same haplotype. $\endgroup$ – Wouter De Coster Dec 3 '18 at 7:11
  • $\begingroup$ In addition, while I don't work in diabetes research I would expect you can find a GWAS with a larger number of individuals than the one you linked. Ideally, you would use a dataset from tens of thousands or more individuals. $\endgroup$ – Wouter De Coster Dec 3 '18 at 7:13
1
$\begingroup$

I think , this kind of research never stop until find way for cure . I have some research in heart disease,and write this paperECG, you can have some research in improve learning method to find fast and more accurate diabetes , there are lot of database your can search in google and more famous is UCI .

| improve this answer | |
$\endgroup$

Not the answer you're looking for? Browse other questions tagged or ask your own question.