Bayesian clinical classification from high-dimensional data: Signatures versus variability

Akram Sharef Said Shalabi; Masato Inoue; Johnathan Watkins; Emanuele de Rinaldis; Anthonius Clara Caspar Coolen

doi:10.1177/0962280216628901

Bayesian clinical classification from high-dimensional data: Signatures versus variability

Akram Sharef Said Shalabi, Masato Inoue, Johnathan Watkins, Emanuele de Rinaldis, Anthonius Clara Caspar Coolen

Research output: Contribution to journal › Article › peer-review

6 Citations (Scopus)

Abstract

When data exhibit imbalance between a large number d of covariates and a small number n of samples, clinical outcome prediction is impaired by overfitting and prohibitive computation demands. Here we study two simple Bayesian prediction protocols that can be applied to data of any dimension and any number of outcome classes. Calculating Bayesian integrals and optimal hyperparameters analytically leaves only a small number of numerical integrations, and CPU demands scale as O(nd). We compare their performance on synthetic and genomic data to the mclustDA method of Fraley and Raftery. For small d they perform as well as mclustDA or better. For d = 10,000 or more mclustDA breaks down computationally, while the Bayesian methods remain efficient. This allows us to explore phenomena typical of classification in high-dimensional spaces, such as overfitting and the reduced discriminative effectiveness of signatures compared to intra-class variability.

Original language	English
Pages (from-to)	336-351
Journal	Statistical Methods in Medical Research
Volume	27
Issue number	2
Early online date	16 Mar 2016
DOIs	https://doi.org/10.1177/0962280216628901
Publication status	Published - 1 Feb 2018

Access to Document

10.1177/0962280216628901

Cite this

@article{b20ac944fc704d7fa175d387baa2f77e,

title = "Bayesian clinical classification from high-dimensional data: Signatures versus variability",

abstract = "When data exhibit imbalance between a large number d of covariates and a small number n of samples, clinical outcome prediction is impaired by overfitting and prohibitive computation demands. Here we study two simple Bayesian prediction protocols that can be applied to data of any dimension and any number of outcome classes. Calculating Bayesian integrals and optimal hyperparameters analytically leaves only a small number of numerical integrations, and CPU demands scale as O(nd). We compare their performance on synthetic and genomic data to the mclustDA method of Fraley and Raftery. For small d they perform as well as mclustDA or better. For d = 10,000 or more mclustDA breaks down computationally, while the Bayesian methods remain efficient. This allows us to explore phenomena typical of classification in high-dimensional spaces, such as overfitting and the reduced discriminative effectiveness of signatures compared to intra-class variability. ",

author = "Shalabi, {Akram Sharef Said} and Masato Inoue and Johnathan Watkins and {de Rinaldis}, Emanuele and Coolen, {Anthonius Clara Caspar}",

year = "2018",

month = feb,

day = "1",

doi = "10.1177/0962280216628901",

language = "English",

volume = "27",

pages = "336--351",

journal = "Statistical Methods in Medical Research",

issn = "0962-2802",

publisher = "SAGE Publications Ltd",

number = "2",

}

TY - JOUR

T1 - Bayesian clinical classification from high-dimensional data

T2 - Signatures versus variability

AU - Shalabi, Akram Sharef Said

AU - Inoue, Masato

AU - Watkins, Johnathan

AU - de Rinaldis, Emanuele

AU - Coolen, Anthonius Clara Caspar

PY - 2018/2/1

Y1 - 2018/2/1

N2 - When data exhibit imbalance between a large number d of covariates and a small number n of samples, clinical outcome prediction is impaired by overfitting and prohibitive computation demands. Here we study two simple Bayesian prediction protocols that can be applied to data of any dimension and any number of outcome classes. Calculating Bayesian integrals and optimal hyperparameters analytically leaves only a small number of numerical integrations, and CPU demands scale as O(nd). We compare their performance on synthetic and genomic data to the mclustDA method of Fraley and Raftery. For small d they perform as well as mclustDA or better. For d = 10,000 or more mclustDA breaks down computationally, while the Bayesian methods remain efficient. This allows us to explore phenomena typical of classification in high-dimensional spaces, such as overfitting and the reduced discriminative effectiveness of signatures compared to intra-class variability.

AB - When data exhibit imbalance between a large number d of covariates and a small number n of samples, clinical outcome prediction is impaired by overfitting and prohibitive computation demands. Here we study two simple Bayesian prediction protocols that can be applied to data of any dimension and any number of outcome classes. Calculating Bayesian integrals and optimal hyperparameters analytically leaves only a small number of numerical integrations, and CPU demands scale as O(nd). We compare their performance on synthetic and genomic data to the mclustDA method of Fraley and Raftery. For small d they perform as well as mclustDA or better. For d = 10,000 or more mclustDA breaks down computationally, while the Bayesian methods remain efficient. This allows us to explore phenomena typical of classification in high-dimensional spaces, such as overfitting and the reduced discriminative effectiveness of signatures compared to intra-class variability.

U2 - 10.1177/0962280216628901

DO - 10.1177/0962280216628901

M3 - Article

SN - 0962-2802

VL - 27

SP - 336

EP - 351

JO - Statistical Methods in Medical Research

JF - Statistical Methods in Medical Research

IS - 2

ER -

Bayesian clinical classification from high-dimensional data: Signatures versus variability

Abstract

Access to Document

Fingerprint

Cite this