Which artificial neural network model can I use when trying to predict gene expression?

Question

I need help choosing a model for my data. I have 50 genes that I need to test for their expression levels. I was thinking of implementing either an SVM or a multilayer perceptron model where the output can be on a scale of 0-1 to show the expression level of each gene. The input data is from single-cell transcriptomic datasets. Can I use either of these models? Which model is better and how? If not, can you please suggest some other models that better fit the data?

Answer 1

You can use any model you want. The answer in machine learning is normally that you have to try each of them out and see how they work; it's usually hard to know in advance what is going to work and what won't. That is the only way to know which will work best.

If you can find published papers in the literature that have tried to solve this problem or a similar problem, you can look at what they did and get some ideas, but if you can't, then you just need to try different approaches and see how well each works.

If you want to predict whether the gene is expressed or not, you are looking for binary classification, and either a SVM or a neural network with a binary softmax output layer.

If you want to predict the degree of expression (from 0-1), then you have a regression problem, and you may need a slightly different technique. You haven't given enough details about the meaning of the output or about the training data you have available, so it's hard to make a recommendation, but one possible option is a neural network with a sigmoid function at the output.