Data‎ > ‎

Prediction of Personality using Diffusion MRI Local Connectome Fingerprints

Local connectome fingerprints (LCF)[1] provide a subject-specific quantification of brain connections in a standard space at the voxel level. The concept is similar to "DNA fingerprinting or DNA profiling", which gives a unique pattern for individuals' genomic characteristics without sequencing the entire genome. LCF, in the similar sense, does not map the entire connectome. It is derived from diffusion MRI to sense microscopic water movement restricted by axons and to capture piece-wise information of the human connectome at the voxel level. LCF allows us to study their connectivity from a bottom-up perspective. 

LCF provides a highly specific measurement for characterizing brain connections and can achieve a uniqueness at 10-6 [1]In comparison, the uniqueness achieved by other fingerprinting approaches such as fMRI or dMRI connectome is only around 1%~10% (error rate, lower the better) [1]. Their low specificity is due to the fact that fMRI is highly sensitive to the functional status of the brain and can be affected by subject's current brain activities, whereas dMRI connectomics uses fiber tracking methods or axonal tracing techniques, which is known to be sensitive to parameters and tends to give a substantial amount of false connections. LCF is based on diffusion MRI and does not rely on fiber tracking to quantify connectivity.

Figure: Local connectome fingerprinting
(A) Local connectome fingerprinting is conducted by first reconstructing diffusion MRI data into a standard space to calculate the spin distribution functions (SDFs). A common fiber direction atlas is then used to sample the density of diffusing water along the fiber directions in the cerebral white matter. The sampled measurements are compiled in a left-posterior-superior order to form a sequence of characteristic values as the local connectome fingerprint. (B) One local connectome fingerprint is shown in different zoom-in resolutions. A local connectome fingerprint has a total of 513,316 entries of scalar values. (C) The local connectome fingerprint of subject #1, #2, and #3 and their repeat measurements (lower row) after 238, 191, and 198 days, respectively. At a coarse level, the local connectome fingerprint differs substantially between three subjects, whereas those from the repeat scans show a remarkably identical pattern, indicating the uniqueness and reproducibility of the local connectome fingerprint.

LCF predicts health and social information

LCF has both the sensitivity and the specificity to be used as a phenotypic marker for subject-specific attributes [2]. LCF is reliable enough that unique phenotype maps can be learned to predict between-subject variability in a range of social, health, and cognitive variables:

Figure: LCF predicts subject's height in a cross-validation study (r=0.5, moderate)

Figure: LCF predicts income in a cross-validation study (r=0.3, weak to moderate)

Figure: LCF weakly predicts in-relation (Is respondent married or in live-in relationship? no = 0; yes = 1)  in a cross-validation study (r=0.14, weak).  

Task: NEO-FFI Prediction

NEO Five-Factor Inventory (NEO-FFI) comprises 60 questionnaires self-reported by the Likert scale (Strongly disagree, Disagree, Neither agree nor disagree,  Agree, Strongly agree) to examine a person's Big Five personality traits (openness to experience, conscientiousness, extraversion, agreeableness, and neuroticism). 

The hypothesis behind this task is that the microscopic connectomics defines fixed behaviour of a subject and thus can be predictive of one's personality.

Data set

There are a total of 1062 subjects included in this data set. Each LCF of a subject has a total of 128894 brain fingerprint features. Each feature has an associated location (mni_location) and fiber orientation (fiber_direction) to allow plotting the spatial distribution of the feature. Please note that each voxel may have more than one feature (because there could be multiple fiber populations within the same voxel).

dimension: the image dimension of the original MRI data (optional)
fiber_direction: the axonal fiber direction for each feature (optional)
mni_location: the spatial location for each feature (optional)
names: HCP serial number for each subject
NEOFAC: subjects answers to 60 questions (variable to be predicted). The NEO-FFI variable can be 0.3: strongly disagree, 0.4 disagree, 0.5 neutral, 0.6: agree, 0.7: strongly agree. 
subjects: The LCFs of 1062 subjects (features)

Download link and allowed-for-distribution license

This Matlab file for the LCF can be downloaded from (531MB)
This file is derived from HCP open access data and shared under term #4 at


Figure: LCF predicts NEO-FFI question 1 (I am not a worrier) with r=0.31

[1] Yeh F-C, Vettel JM, Singh A, Poczos B, Grafton ST, Erickson KI, et al. (2016) Quantifying Differences and Similarities in Whole-Brain White Matter Architecture Using Local Connectome Fingerprints. PLoS Comput Biol 12(11): e1005203.
[2] Powell, M. A., Garcia, J. O., Yeh, F. C., Vettel, J. M., & Verstynen, T. (2018). Local connectome phenotypes predict social, health, and cognitive factors. Network Neuroscience, 2(1), 86-105.