Skip to content
PaperFren

Can a CNN predict how cancer cells respond to drugs?

Open paper intelligence

A two-branch convolutional network predicted drug response well for drugs it had seen before, but almost failed on entirely new drugs.

Source

Improving prediction of phenotypic drug response on cancer cell lines using deep convolutional network

Liu P, Li H, Li S, et al. · BMC bioinformatics · 2019

doi.org/10.1186/s12859-019-2910-6Read the full paper ↗114 citationscc by

Study at a glance

Design
Computational / modelling — Regression of normalised IC50 from one-hot SMILES and 735 genetic features; random 80/10/10 pair splits repeated 50 times, plus drug-blind, cell-line-blind and tissue-blind splits.
N
N=172114 · 172,114 measured drug-cell line IC50 values among 223 drugs and 948 cell lines from GDSC.
Population
Anti-cancer drugs and cancer cell lines in the Genomics of Drug Sensitivity in Cancer database
Outcome
Coefficient of determination, Pearson correlation and RMSE between predicted and measured IC50

Structured fields used in claim comparison tables when every cited study has a complete layer.

What they did

The authors built tCNNS, which runs one 1D convolutional network over a drug's SMILES string (encoded as a one-hot matrix) and another over a cell line's mutation and copy-number features, then combines them in fully connected layers to predict the drug's IC50. They used 172,114 measured drug-cell line pairs covering 223 drugs and 948 cell lines. Beyond random splits, they tested on held-out drugs, held-out cell lines and held-out tissues, and scrambled the SMILES input to see what the network actually learned.

What they found

With random pair splits the model reached a mean R squared of 0.826 and Pearson correlation of 0.909 over 50 repeats, above an older neural network baseline. When whole drugs were held out, performance collapsed (mean Pearson correlation around 0.2), while held-out cell lines were handled much better. Shuffling or rearranging the SMILES barely changed accuracy, showing the network was exploiting statistical patterns rather than chemical meaning; it also stayed accurate with far fewer cell-line features.

The limits

What it doesn't show

The comparison with the earlier baseline is indirect because the database version changed, which the authors say may be overly optimistic. Random pair splits let the same drugs and cell lines appear in training and testing, so the headline scores mainly reflect filling in a table, not predicting new drugs. Suggested new drug uses were supported only by literature searches, not lab tests, and the scrambling result means the model offers little chemical insight.

Key terms

IC50
The drug concentration needed to inhibit cancer cell growth by half; lower means more potent.
SMILES
A text string that encodes a molecule's atoms and bonds in a line.
One-hot encoding
Representing each symbol as a vector with a single 1 in its own position and 0s elsewhere.
Blind test
An evaluation where whole drugs or cell lines are kept out of training, so the model must generalise to unseen entities.
Coefficient of determination (R squared)
The proportion of variance in the true values that the predictions explain.

Flashcards

1 / 9

0 of 9 answers reviewed

Research intelligence for this paper

See its role on concept claims, tensions it is part of, placement history, and related discoveries.

Open paper intelligence

Quiz yourself

1 / 5

What does tCNNS predict?

Common questions

Why does a random split give such high scores?

Each drug and cell line appears in training with other partners, so the model can learn how potent each drug and how sensitive each cell line is on average.

What does the SMILES-scrambling experiment show?

Accuracy stayed about the same even when the chemical meaning was destroyed, so the network was mainly identifying drugs, not learning chemistry.

Is the model useful for brand-new drugs?

Not yet: in the drug-blind test R squared fell to barely above zero.

More on Convolutional networks