This website requires JavaScript.

Latent Similarity Identifies Important Functional Connections for Phenotype Prediction

Anton OrlichenkoGang QuGemeng Zhang ...+4 Yu-Ping Wang
Aug 2022
摘要
Objective: Endophenotypes like brain age and fluid intelligence are importantbiomarkers of disease status. However, brain imaging studies to identify thesebiomarkers often encounter limited numbers of subjects and high dimensionalimaging features, hindering reproducibility. Therefore, we develop aninterpretable, multivariate classification/regression algorithm, called LatentSimilarity (LatSim), suitable for small sample size, high feature dimensiondatasets. Methods: LatSim combines metric learning with a kernel similarityfunction and softmax aggregation to identify task-related similarities betweensubjects. Inter-subject similarity is utilized to improve performance on threeprediction tasks using multi-paradigm fMRI data. A greedy selection algorithm,made possible by LatSim's computational efficiency, is developed as aninterpretability method. Results: LatSim achieved significantly higherpredictive accuracy at small sample sizes on the PhiladelphiaNeurodevelopmental Cohort (PNC) dataset. Connections identified by LatSim gavesuperior discriminative power compared to those identified by other methods. Weidentified 4 functional brain networks enriched in connections for predictingbrain age, sex, and intelligence. Conclusion: We find that most information fora predictive task comes from only a few (1-5) connections. Additionally, wefind that the default mode network is over-represented in the top connectionsof all predictive tasks. Significance: We propose a novel algorithm for smallsample, high feature dimension datasets and use it to identify connections intask fMRI data. Our work should lead to new insights in both algorithm designand neuroscience research.
展开全部
图表提取

暂无人提供速读十问回答

论文十问由沈向洋博士提出,鼓励大家带着这十个问题去阅读论文,用有用的信息构建认知模型。写出自己的十问回答,还有机会在当前页面展示哦。

Q1论文试图解决什么问题?
Q2这是否是一个新的问题?
Q3这篇文章要验证一个什么科学假设?
0
被引用
笔记
问答