What inputs are required for what outputs with machine learning algorithms?

I am a biologist, relatively new to software development as a whole. Our lab has some ideas on inputting positive and negative experimental results into a machine learning algorithm with the goal of finding what exactly makes the results turn out a particular way.

We have algorithms that design siRNAs, and want to improve them based on experimental data and metadata.

Specifically, the goal is to find out how the structure of a genomic locus influences the ability of designed siRNAs to silence the particular gene on that locus. Subsequently, we want to be able to make better predictions about the siRNA designs that would be required to optimally silence that gene.

The big question is what kind of data do we feed the algorithm, and then how does it handle that data and turn it into something useful? How does the input data result in a better siRNA design algorithm? What steps are involved?

Any help on this would be super useful for helping us understand what expertise we need to pull in for a grant, and plan our work for the next few years.

Thanks!