Back

The LMSz method - an automatable scalable approach to constructing gene-specific growth charts in rare disorders.

Low, K. J.; Foreman, J.; Hobson, R.; Kwuo, H.; Martinez-Cayuelas, E.; Almoguera Castillo, B.; Marin-Reina, P.; Caraffi, S. G.; Garavelli, L.; Woods, E.; Balasubramanian, M.; Bayat, A.; Ockeloen, C. W.; Wright, C. M.; Firth, H. V.; Cole, T. J.

2024-08-20 genetic and genomic medicine
10.1101/2024.08.19.24312213 medRxiv
Show abstract

IntroductionChildren with monogenic neurodevelopmental disorders often grow abnormally. Gene-specific growth charts would be useful but require large samples to construct them using the conventional LMS method. MethodsWe transformed anthropometry to British 1990 reference z-scores for 328 UK and 264 international probands with ANKRD11, ARID1B, ASXL3, DDX3X, KMT2A or SATB2-related disorders, and modelled mean and standard deviation (SD) of the z-scores as gene-specific linear age trends adjusted for sex. Back-transforming the mean {+/-}2 SD lines provided gene-specific median, 2nd and 98th centiles. ResultsThe resulting z-score charts look plausible on several counts. Only KMT2A shows a (rising) age trend in median height, while BMI and weight increase in several genes, possibly reflecting population trends. Apart from SATB2 and DDX3X, the gene-specific medians are all below the reference (range 0.1th centile for height KMT2A to 36th centile for BMI ANKRD11). Median OFC shows no age trend, with medians ranging from 10th-30th centile, and ASXL3 lowest, on the 3rd centile. There are no sex differences in 19/24 cases. ConclusionsOur LMSz method produces gene-specific growth charts for rare diseases, an essential clinical tool for paediatric care. We plan to automate it within the DECIPHER platform, enabling availability for all relevant genes.

Published in European Journal of Human Genetics (predicted rank #1) · training set

Matching journals

The top 6 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.