School of Informatics


Edinburgh researchers provide data analysis for a cancer study that revealed weak spots in chemotherapy-resistant cells

Diego Oyarzún and doctoral student Denise Thiel are co-authors of a PNAS paper led by Imperial College scientists that explores ‘weak spots' in cancer cells recovering from chemotherapy.

Figure 1E as described in the text
Figure 1E (PNAS, DOI: 10.1073/pnas.2018229118)

Cancers are notoriously difficult to treat because they have the ability to withstand the toxic effects of chemotherapy, making therapies for extremely invasive cancers ineffective.

This study looked at the effects of chemotherapy on multiple myeloma cells, an incurable form of bone marrow cancer. The findings reveal that cells recovering from chemotherapy show specific vulnerabilities which could be targeted by a ‘second punch’ treatment. Researchers suggest that GCN2 – a specialised enzyme that senses and responds to amino acid deficiency – could be an ‘Achilles’ heel’ for recovering myeloma cells.

The Edinburgh team supported the machine learning by combining Gaussian process regression and graph diffusion processes to carry out unsupervised clustering of large, high-dimensional, time series data.

Their contribution is summarized in Figures 1E-F of the paper. As a starting point, they were given ~15k time series (one series per gene in the cell), with each one having 7 time points and 5 repeats. Such a huge amount of data would be difficult to analyse manually.

To resolve this, they converted the data into the network graph of Fig 1E, which condenses all that huge data into a compact and visually clear form. In Fig 1E, each dot is a gene in the cell, and the connections between dots represent how similar their temporal responses are.

Figure 1F as described in the text
Figure 1F (PNAS, DOI: 10.1073/pnas.2018229118)

In the next step they applied community detection algorithms to the network in Fig 1E to discover the six dominant ‘patterns’ of responses, which are shown in Fig 1F.

Each pattern depicts how genes respond to the drug treatment, for example, in Pattern 1 in Fig 1F, genes ‘go’ up and then down, whereas in Pattern 5 they ‘go’ down and then up.

A key finding is that one of those patterns revealed a new vulnerability in cells. This vulnerability can potentially be targeted with a second drug which will prevent cells from recovering.  As these are cancer cells, effective treatment must make sure they don’t recover.

The paper ‘Systems level profiling of chemotherapy-induced stress resolution in cancer cells reveals druggable trade-offs’ by Paula Saavedra-García et al is now available to read from PNAS.

Related links

Full paper in PNAS

Diego's personal page