MultiPLIER: a transfer learning framework for transcriptomics reveals systemic features of rare disease

Abstract

SUMMARYUnsupervised machine learning methods provide a promising means to analyze and interpret large datasets. However, most datasets generated by individual researchers remain too small to fully benefit from these methods. In the case of rare diseases, there may be too few cases available, even when multiple studies are combined. We sought to determine whether or not machine learning models could be constructed from large public data compendia and then transferred to small datasets for subsequent analysis. We trained models using Pathway Level Information ExtractoR (PLIER) over datasets of different types and scales. Models constructed from large public datasets were i) more detailed than those constructed from individual datasets; ii) included features that aligned well to important biological factors; iii) transferrable to rare disease datasets where the models describe biological processes related to disease severity more effectively than models trained within those datasets.

Differential Principal Components Reveal Patterns of Differentiation in Case/Control Studies

“…We are interested to see the variety of unsupervised dimensionality reduction techniques that can be repurposed in this supervised setting. For instance, we may want to estimate components which correspond to genetic pathways, similarly to (27,28). This can be accomplished under the differential PCA framework by using a Bayesian PCA method with a prior that links genes according to pathway annotations.…”

Integrative analysis identifies candidate tumor microenvironment and intracellular signaling pathways that define tumor heterogeneity in NF1

“…We retrieved a model [41] that was previously trained on the recount2 RNA-seq dataset [25,42] and then used it to assess the expression of latent variables in the pan-NF dataset. We retrieved code for this analysis from the public repository for MultiPLIER [27] (https://github.com/greenelab/multi-plier). To project the NF data into the MultiPLIER model, we used the GetNewDataB function.…”

“…We analyzed transcriptomic data from NF in the context of latent variables from MultiPLIER, a machine learning resource designed to aid in rare disease analyses [27] . Raw transcriptomic data from NF were retrieved from the NF Data Portal, reprocessed, and stored in Synapse as described above.…”

“…Given the low sample size compared to the large feature space of RNA-seq data, we applied a transfer learning-inspired approach to meaningfully reduce the feature space with minimal decrease in information content. Transfer learning techniques can leverage large well curated datasets such as recount2 [25] to identify latent variables (LVs): groups of genes that show common transcriptomic patterns that are relevant to a specific subset of samples [26,27] . We transferred a machine learning model [27] trained on recount2 to assess LV expression in the NF1 nerve sheath tumor dataset.…”

“…Transfer learning techniques can leverage large well curated datasets such as recount2 [25] to identify latent variables (LVs): groups of genes that show common transcriptomic patterns that are relevant to a specific subset of samples [26,27] . We transferred a machine learning model [27] trained on recount2 to assess LV expression in the NF1 nerve sheath tumor dataset. While many of these LVs map to known signatures (i.e.…”