A recent study conducted at Uppsala University analyzed neural networks previously used to understand social networks and applied it to analyze mRNA production in human tissue.
Currently, the most common method to analyzing the presence of mRNA in a tissue at the microscopic level is using in situ sequencing. This method requires a lot of manual labelling and identifying of mRNA, cell types and tissue to allow for any substantial analysis to occur.
A good practice when building machine learning models is to ensure the model does not overfit to the data being presented to it. Overfitting refers to the concept whereby a neural network is trained with a certain set of data almost perfectly, but is unable to predict new data as it cannot generalize well. Keeping this in mind, many machine learning researchers will attempt to build models that can generalize to any data set.
The researchers of the study used a model previously analyzing social networks. The model identified clusters of individuals based on similar followers on Twitter, similar Google Searches and many more similarities and differences on the internet.
When the model was provided with the cell data (images of human tissue with dots marking mRNA presence), the model was able to correctly cluster different tissues, cell types and identify the mRNA markers. Seeing success from models like this is a huge breakthrough in mapping cell types and mRNA functions.
Not only does this help scientists within this field better understand their research, it also depicts how networks can be generalized to help understand various different applications. A map of social network similarities is able to map cell tissue today, who knows what we will be able to map in the future.
Sources:
https://www.sciencedaily.com/releases/2020/10/201019111918.htm