Functional Magnetic Resonance Imaging (fMRI) provides means to characterize brain activations in response to behavior. However, cognitive neuroscience has been limited to group-level effects referring to the performance of specific tasks. To obtain the functional profile of elementary cognitive mechanisms, the combination of brain responses to many tasks is required, by pooling data or results from different single-task studies. Meta-analyses allow the accumulation of knowledge across studies. Yet, they are typically impacted not only by inter-subject and inter-site variability but also loss of information from sparse peak-coordinate representations. In this talk, I will address a battery of studies, which combine deep phenotyping and multitask-fMRI approaches to extensively investigate the functional signatures of the different components that characterize the human behavior. First, I will describe a set of experiments, based on temporally controlled task designs, reported in [1], [2] and [3], in which we leverage a collection of source task-fMRI data from the Individual Brain Charting (IBC) dataset. The main goal herein is to investigate the feasibility of performing individual functional brain atlasing, free from inter-subject and inter-site variability, as an effort to establish an univocal relationship between functional segregation of brain regions and elementary mental functions. Results show that individual topographies---common to all tasks---are consistently mapped within and, to a lesser extent, across participants. Besides, prediction scores associated with the reconstruction of contrasts of one task from the remaining ones reveal the quantitative contribution of each task to these common representations. Yet, scores decreased when subjects were permuted between train and test, confirming that topographies are driven by subject-specific variability. In addition, we demonstrate how cognitive mapping can benefit from contrasts accumulation, by analyzing the functional fingerprints of a set of individualized regions-of-interest from the language network. Second, I will describe our ongoing work on the quality-assessment and validation of a subset of tasks from the IBC dataset based on naturalistic stimuli using two types of encoding models: the unsupervised Fast Shared Response Model [4], and a feature-defined model based on Deep Convolutional Neural Networks [5, 6]. [1] Pinho, A.L. et al. (2021) DOI: 10.1002/hbm.25189 [2] Pinho, A.L. et al. (2018) DOI: 10.1038/sdata.2018.105 [3] Pinho, A.L. et al. (2020) DOI: 10.1038/s41597-020-00670-4 [4] Richard, H. et al. (2019) DOI: 10.48550/arXiv.1909.12537 [5] Eickenberg, M. et al. (2016) DOI: 10.1016/j.neuroimage.2016.10.001 [6] Güçlü, U. and van Gerven, M. A. J. (2015) DOI: 10.1523/JNEUROSCI.5023-14.2015