Overcoming limitations in current measures of drug response may enable AI-driven precision oncology.

Machine learning (ML) models of drug sensitivity prediction are becoming increasingly popular in precision oncology. Here, we identify a fundamental limitation in standard measures of drug sensitivity that hinders the development of personalized prediction models - they focus on absolute effects but do not capture relative differences between cancer subtypes. Our work suggests that using z-scored drug response measures mitigates these limitations and leads to meaningful predictions, opening the door for sophisticated ML precision oncology models.

SpaceM reveals metabolic states of single cells.

A growing appreciation of the importance of cellular metabolism and revelations concerning the extent of cell-cell heterogeneity demand metabolic characterization of individual cells. We present SpaceM, an open-source method for in situ single-cell metabolomics that detects >100 metabolites from >1,000 individual cells per hour, together with a fluorescence-based readout and retention of morpho-spatial features. We validated SpaceM by predicting the cell types of cocultured human epithelial cells and mouse fibroblasts. We used SpaceM to show that stimulating human hepatocytes with fatty acids leads to the emergence of two coexisting subpopulations outlined by distinct cellular metabolic states. Inducing inflammation with the cytokine interleukin-17A perturbs the balance of these states in a process dependent on NF-κB signaling. The metabolic state markers were reproduced in a murine model of nonalcoholic steatohepatitis. We anticipate SpaceM to be broadly applicable for investigations of diverse cellular models and to democratize single-cell metabolomics.

OffsampleAI: artificial intelligence approach to recognize off-sample mass spectrometry images.

BACKGROUND: Imaging mass spectrometry (imaging MS) is an enabling technology for spatial metabolomics of tissue sections with rapidly growing areas of applications in biology and medicine. However, imaging MS data is polluted with off-sample ions caused by sample preparation, particularly by the MALDI (matrix-assisted laser desorption/ionization) matrix application. Off-sample ion images confound and hinder statistical analysis, metabolite identification and downstream analysis with no automated solutions available. RESULTS: We developed an artificial intelligence approach to recognize off-sample ion images. First, we created a high-quality gold standard of 23,238 expert-tagged ion images from 87 public datasets from the METASPACE knowledge base. Next, we developed several machine and deep learning methods for recognizing off-sample ion images. The following methods were able to reproduce expert judgements with a high agreement: residual deep learning (F1-score 0.97), semi-automated spatio-molecular biclustering (F1-score 0.96), and molecular co-localization (F1-score 0.90). In a test-case study, we investigated off-sample images corresponding to the most common MALDI matrix (2,5-dihydroxybenzoic acid, DHB) and characterized properties of matrix clusters. CONCLUSIONS: Overall, our work illustrates how artificial intelligence approaches enabled by open-access data, web technologies, and machine and deep learning open novel avenues to address long-standing challenges in imaging MS.

ColocML: machine learning quantifies co-localization between mass spectrometry images.

MOTIVATION: Imaging mass spectrometry (imaging MS) is a prominent technique for capturing distributions of molecules in tissue sections. Various computational methods for imaging MS rely on quantifying spatial correlations between ion images, referred to as co-localization. However, no comprehensive evaluation of co-localization measures has ever been performed; this leads to arbitrary choices and hinders method development. RESULTS: We present ColocML, a machine learning approach addressing this gap. With the help of 42 imaging MS experts from nine laboratories, we created a gold standard of 2210 pairs of ion images ranked by their co-localization. We evaluated existing co-localization measures and developed novel measures using term frequency-inverse document frequency and deep neural networks. The semi-supervised deep learning Pi model and the cosine score applied after median thresholding performed the best (Spearman 0.797 and 0.794 with expert rankings, respectively). We illustrate these measures by inferring co-localization properties of 10 273 molecules from 3685 public METASPACE datasets. AVAILABILITY AND IMPLEMENTATION: https://github.com/metaspace2020/coloc. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.