Beyond attention – new insights into the neural basis of intelligence and cognitive abilities through machine learning-based predictive modeling approaches

智力对教育、职业、甚至健康和寿命都具有预测意义。尽管智力在人类大脑的结构和功能中具有生物学基础的假设已经相对确立，但对这一基础的全面理解仍然缺乏，但这是正在进行的研究的一个重要目标。将基于机器学习的预测建模方法引入神经科学，以及网络分析的进步和大型神经成像数据集的发布，为从新的角度解决这一目标提供了新的机会。在先前项目公布的结果的基础上，该项目揭示了大脑网络重构是一种有前途的智能生物标志物，本文提出的后续项目将把重点扩大到：a)预测个体智力分数，而不是事后解释方差（预测建模，而不是相关性分析）；b)识别智能预测网络指纹（大脑区域之间的通信模式）；c)预测智力的神经特征是如何在人脑中实现的问题。为了实现这些目标，我们将首先将建立的基于连接体的预测建模（CPM）方法和我们的研究小组开发的基于协方差最大化特征向量的预测（CMEP）方法应用于来自三个大样本（N = 806来自人类连接体项目；N = 138和N = 184来自阿姆斯特丹开放MRI收集）的功能磁共振成像（fMRI）数据，以验证以下假设：大脑网络重构可以显著预测个体智力得分。2. 不同的大脑网络对智力预测的贡献差别很大。最后，在计划项目的最后一部分，将实施神经网络，目的是测试以下假设：结合神经功能的不同特征，可以显著提高对智力的预测。这些研究问题的答案将增强我们对人类智力的神经生物学基础的理解，并从神经生物学的角度为智力和认知次级能力之间的假设关系提供心理学理论——这些见解也有助于我们对认知障碍疾病的理解。最后，开发的方法将用于未来的神经科学研究。

Intelligence has predictive relevance for education, occupation, and even for health and longevity. Although the assumption that intelligence has a biological basis within the structure and function of the human brain is relatively established, a comprehensive understanding of this basis is still lacking but constitutes an important aim of ongoing research. The introduction of machine learning-based predictive modeling approaches to neuroscience together with advances in network analyses and the release of large neuroimaging data sets opens new opportunities to address this aim from a new perspective. Building on the published results of the previous project, which revealed brain network reconfiguration as a promising biomarker of intelligence, the here proposed follow-up project will broaden the focus to a) the prediction of individual intelligence scores instead of explaining variance post-hoc (predictive modeling instead of correlation analyses), b) the identification of intelligence-predictive network fingerprints (communication patterns between brain regions), and c) the question of how intelligence-predictive neural characteristics are implemented within the human brain. To achieve these goals, we will first apply the established connectome-based predictive modeling (CPM) approach and the, in our research group developed, covariance maximizing eigenvector-based prediction (CMEP) methodology to functional magnetic resonance imaging (fMRI) data from three large samples (N = 806 from the Human Connectome Project; N = 138 and N = 184 from the Amsterdam Open MRI Collection) to test the following hypotheses: 1. Individual intelligence scores can be significantly predicted from brain network reconfiguration. 2. The contribution of different brain networks to the prediction of intelligence differs significantly. Finally, in the last part of the planned project, Neural Networks will be implemented with the aim of testing the following hypothesis: 3. The prediction of intelligence can be significantly improved by combining different features of neural functioning. Answers to these research questions will enhance our understanding about the neurobiological basis of human intelligence and inform psychological theories about postulated relationships between intelligence and cognitive sub-abilities from a neurobiological perspective - insights which also contribute to our understanding about diseases with cognitive impairments. Finally, the developed methodology will be made available for future neuroscientific research.