Brain age prediction using deep learning uncovers associated sequence variants

<jats:title>Abstract</jats:title><jats:p>Machine learning algorithms can be trained to estimate age from brain structural MRI. The difference between an individual’s predicted and chronological age, predicted age difference (PAD), is a phenotype of relevance to aging and brain disease. Here, we present a new deep learning approach to predict brain age from a T1-weighted MRI. The method was trained on a dataset of healthy Icelanders and tested on two datasets, IXI and UK Biobank, utilizing transfer learning to improve accuracy on new sites. A genome-wide association study (GWAS) of PAD in the UK Biobank data (discovery set: <jats:inline-formula><jats:alternatives><jats:tex-math>$$N=12378$$</jats:tex-math><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mi>N</mml:mi> <mml:mo>=</mml:mo> <mml:mn>12378</mml:mn> </mml:math></jats:alternatives></jats:inline-formula>, replication set: <jats:inline-formula><jats:alternatives><jats:tex-math>$$N=4456$$</jats:tex-math><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mi>N</mml:mi> <mml:mo>=</mml:mo> <mml:mn>4456</mml:mn> </mml:math></jats:alternatives></jats:inline-formula>) yielded two sequence variants, rs1452628-T (<jats:inline-formula><jats:alternatives><jats:tex-math>$$\beta =-0.08$$</jats:tex-math><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mi>β</mml:mi> <mml:mo>=</mml:mo> <mml:mo>−</mml:mo> <mml:mn>0.08</mml:mn> </mml:math></jats:alternatives></jats:inline-formula>, <jats:inline-formula><jats:alternatives><jats:tex-math>$$P=1.15\times{10}^{-9}$$</jats:tex-math><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mi>P</mml:mi> <mml:mo>=</mml:mo> <mml:mn>1.15</mml:mn> <mml:mo>×</mml:mo> <mml:msup> <mml:mrow> <mml:mn>10</mml:mn> </mml:mrow> <mml:mrow> <mml:mo>−</mml:mo> <mml:mn>9</mml:mn> </mml:mrow> </mml:msup> </mml:math></jats:alternatives></jats:inline-formula>) and rs2435204-G (<jats:inline-formula><jats:alternatives><jats:tex-math>$$\beta =0.102$$</jats:tex-math><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mi>β</mml:mi> <mml:mo>=</mml:mo> <mml:mn>0.102</mml:mn> </mml:math></jats:alternatives></jats:inline-formula>, <jats:inline-formula><jats:alternatives><jats:tex-math>$$P=9.73\times 1{0}^{-12}$$</jats:tex-math><mml:math xmlns:mml="http://www.w3.org/1998/Math/MathML"> <mml:mi>P</mml:mi> <mml:mo>=</mml:mo> <mml:mn>9.73</mml:mn> <mml:mo>×</mml:mo> <mml:mn>1</mml:mn> <mml:msup> <mml:mrow> <mml:mn>0</mml:mn> </mml:mrow> <mml:mrow> <mml:mo>−</mml:mo> <mml:mn>12</mml:mn> </mml:mrow> </mml:msup> </mml:math></jats:alternatives></jats:inline-formula>). The former is near <jats:italic>KCNK2</jats:italic> and correlates with reduced sulcal width, whereas the latter correlates with reduced white matter surface area and tags a well-known inversion at 17q21.31 (H2).</jats:p>