Bear M Goldstein, Agnieszka Pluta, Grace Q Miao, Ashley L Binnquist, Matthew D Lieberman
{"title":"多时间点模式分析(MTPA):利用神经时间序列数据改进分类。","authors":"Bear M Goldstein, Agnieszka Pluta, Grace Q Miao, Ashley L Binnquist, Matthew D Lieberman","doi":"10.1093/scan/nsaf058","DOIUrl":null,"url":null,"abstract":"<p><p>Long, naturalistic stimuli are effective in evoking meaningfully differential neural response patterns between groups. However, the resulting timeseries data often have a high number of features compared to a limited sample size, increasing the likelihood of overfitting and reducing predictive power. This paper introduces multi-timepoint pattern analysis (MTPA) as a temporal dimension reduction approach for improving prediction accuracy when building models with long neural timeseries data. Using feature selection with elastic net regression, MTPA identifies predictive neural patterns while preserving the temporal structure and interpretability of the data. Across two experiments with distinct populations and objectives, MTPA demonstrated consistent advantages over approaches using principal component analysis, windowed averaging, and no dimension reduction. Experiment 1 predicted persistent work-related psychological states in business professionals, achieving accuracies up to 79.1%. Experiment 2 predicted cognitive load and narrative context during video viewing in undergraduates, with accuracies up to 66.5%. These findings suggest that MTPA may be a useful tool for analyzing neural data from extended naturalistic designs, enabling researchers to improve prediction accuracy across diverse outcomes and obtain new insights into the temporal dynamics of neural responses.</p>","PeriodicalId":94208,"journal":{"name":"Social cognitive and affective neuroscience","volume":" ","pages":""},"PeriodicalIF":0.0000,"publicationDate":"2025-06-11","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":"{\"title\":\"Multi-timepoint pattern analysis (MTPA): Improving classification with neural timeseries data.\",\"authors\":\"Bear M Goldstein, Agnieszka Pluta, Grace Q Miao, Ashley L Binnquist, Matthew D Lieberman\",\"doi\":\"10.1093/scan/nsaf058\",\"DOIUrl\":null,\"url\":null,\"abstract\":\"<p><p>Long, naturalistic stimuli are effective in evoking meaningfully differential neural response patterns between groups. However, the resulting timeseries data often have a high number of features compared to a limited sample size, increasing the likelihood of overfitting and reducing predictive power. This paper introduces multi-timepoint pattern analysis (MTPA) as a temporal dimension reduction approach for improving prediction accuracy when building models with long neural timeseries data. Using feature selection with elastic net regression, MTPA identifies predictive neural patterns while preserving the temporal structure and interpretability of the data. Across two experiments with distinct populations and objectives, MTPA demonstrated consistent advantages over approaches using principal component analysis, windowed averaging, and no dimension reduction. Experiment 1 predicted persistent work-related psychological states in business professionals, achieving accuracies up to 79.1%. Experiment 2 predicted cognitive load and narrative context during video viewing in undergraduates, with accuracies up to 66.5%. These findings suggest that MTPA may be a useful tool for analyzing neural data from extended naturalistic designs, enabling researchers to improve prediction accuracy across diverse outcomes and obtain new insights into the temporal dynamics of neural responses.</p>\",\"PeriodicalId\":94208,\"journal\":{\"name\":\"Social cognitive and affective neuroscience\",\"volume\":\" \",\"pages\":\"\"},\"PeriodicalIF\":0.0000,\"publicationDate\":\"2025-06-11\",\"publicationTypes\":\"Journal Article\",\"fieldsOfStudy\":null,\"isOpenAccess\":false,\"openAccessPdf\":\"\",\"citationCount\":\"0\",\"resultStr\":null,\"platform\":\"Semanticscholar\",\"paperid\":null,\"PeriodicalName\":\"Social cognitive and affective neuroscience\",\"FirstCategoryId\":\"1085\",\"ListUrlMain\":\"https://doi.org/10.1093/scan/nsaf058\",\"RegionNum\":0,\"RegionCategory\":null,\"ArticlePicture\":[],\"TitleCN\":null,\"AbstractTextCN\":null,\"PMCID\":null,\"EPubDate\":\"\",\"PubModel\":\"\",\"JCR\":\"\",\"JCRName\":\"\",\"Score\":null,\"Total\":0}","platform":"Semanticscholar","paperid":null,"PeriodicalName":"Social cognitive and affective neuroscience","FirstCategoryId":"1085","ListUrlMain":"https://doi.org/10.1093/scan/nsaf058","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"","JCRName":"","Score":null,"Total":0}
Multi-timepoint pattern analysis (MTPA): Improving classification with neural timeseries data.
Long, naturalistic stimuli are effective in evoking meaningfully differential neural response patterns between groups. However, the resulting timeseries data often have a high number of features compared to a limited sample size, increasing the likelihood of overfitting and reducing predictive power. This paper introduces multi-timepoint pattern analysis (MTPA) as a temporal dimension reduction approach for improving prediction accuracy when building models with long neural timeseries data. Using feature selection with elastic net regression, MTPA identifies predictive neural patterns while preserving the temporal structure and interpretability of the data. Across two experiments with distinct populations and objectives, MTPA demonstrated consistent advantages over approaches using principal component analysis, windowed averaging, and no dimension reduction. Experiment 1 predicted persistent work-related psychological states in business professionals, achieving accuracies up to 79.1%. Experiment 2 predicted cognitive load and narrative context during video viewing in undergraduates, with accuracies up to 66.5%. These findings suggest that MTPA may be a useful tool for analyzing neural data from extended naturalistic designs, enabling researchers to improve prediction accuracy across diverse outcomes and obtain new insights into the temporal dynamics of neural responses.