Skip to content
PaperFren

Can wiring a neural net by biology help small-data prognosis?

Open paper intelligence

A neural network whose first connections follow known gene pathways, and which prunes itself to stay sparse, predicted brain-tumour survival slightly better than standard classifiers while being easier to interpret.

Source

PASNet: pathway-associated sparse deep neural network for prognosis prediction from high-throughput data

Hao J, Kim Y, Kim TK, et al. · BMC bioinformatics · 2018

doi.org/10.1186/s12859-018-2500-zRead the full paper ↗91 citationscc by

Study at a glance

Design
Computational / modelling — New pathway-masked sparse neural network vs four baseline classifiers, stratified 5-fold cross-validation repeated ten times on TCGA glioblastoma gene expression.
N
N=475 · Patient samples after excluding censored cases: 99 long-term survivors (past 24 months) and 376 short-term survivors; 4359 genes in 574 Reactome pathways as inputs.
Population
Glioblastoma patients' tumour gene-expression profiles from The Cancer Genome Atlas
Outcome
Classification of long-term vs short-term survival, measured by AUC and F1-score

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

What they did

The authors built PASNet, a neural network with a gene input layer, a pathway layer connected only to each pathway's member genes, one hidden layer and an output layer. It was trained by repeatedly training small dropout sub-networks, pruning weak weights at a per-layer sparsity level, and weighting errors by class to handle imbalance. Using glioblastoma gene expression from TCGA, they predicted whether patients survived beyond 24 months and compared PASNet with an SVM, random LASSO, LASSO logistic regression and a same-size fully connected dropout network.

What they found

PASNet had the highest mean AUC (0.6622) and F1-score (0.3978) across the repeated cross-validation, ahead of the dropout network (AUC 0.6408) and SVM (0.6337), and the authors report these differences as statistically significant. Nonlinear models beat the linear LASSO models. Inspecting the trained weights highlighted pathways such as GPCR signalling and the innate immune system, many of which the authors note have prior links to glioblastoma.

The limits

What it doesn't show

Absolute accuracy is modest: an AUC of 0.6622 means the model separates survivors only somewhat better than chance, and the gains over baselines are a couple of points. Everything was evaluated on one cancer dataset with no external validation cohort, and hyperparameters were tuned separately for PASNet and the dense network by hand-guided search. The biological interpretation is post hoc; the authors themselves caution that a node being active in the network does not mean the pathway is biologically activated.

Key terms

High-dimension, low-sample size (HDLSS)
Data with far more features (here thousands of genes) than examples (hundreds of patients), which makes overfitting easy.
Sparse coding / pruning
Setting many network weights to zero so only a few important connections remain, which reduces overfitting and makes the model easier to read.
Pathway layer
A layer where each node stands for a biological pathway and connects only to genes known to belong to it.
AUC
Area under the ROC curve: the probability a model ranks a random positive case above a random negative one; 0.5 is chance.
Cost-sensitive learning
Weighting errors so mistakes on the rarer class count as much as mistakes on the common class.

Flashcards

1 / 10

0 of 10 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 fixes the connections between PASNet's gene and pathway layers?

Common questions

Why not just use a normal fully connected network?

With thousands of genes and only hundreds of patients, a dense network has too many weights to fit reliably and its hidden units are hard to interpret; the pathway wiring builds in prior knowledge and cuts parameters.

Is an AUC of 0.66 good?

It is better than chance and than the baselines here, but it is still weak for clinical use; the paper's stronger claim is interpretability, not accuracy.

Do the highlighted pathways prove what drives survival?

No. They are patterns in the trained weights that the authors compare with prior literature; they are hypotheses needing experimental confirmation.

More on Interpretability