[Effect of RNA Sequencing Normalization on Forensic Age Estimation]

To evaluate the applicability of four commonly used RNA sequencing normalization methods in forensic age estimation, and to provide a reference for optimizing the accuracy and precision of age estimation models.

Peripheral blood samples were collected from 147 unrelated Chinese Han individuals. After RNA sequencing, gene expression levels were normalized using four methods: counts per million (CPM), fragments per kilobase of transcript per million mapped reads (FPKM), transcripts per million (TPM), and trimmed mean of M-values (TMM). Spearman correlation analysis was used to screen age-related mRNAs. Three age estimation models were constructed using least absolute shrinkage and selection operator (LASSO), support vector machine (SVM), and extreme gradient boosting (XGBoost), respectively, to compare the performance of the four normalization methods.

The four normalization methods screened different numbers of potential age-related markers. The TMM method screened the largest number of 2 912 markers. The number of potential markers screened was 2 897 and 1 481 for CPM and FPKM, respectively. The TPM method screened the fewest markers with the number of 338. Age estimation models constructed based on these markers exhibited mean absolute errors (MAE) ranging from 4.38 to 8.62 years in the training set and from 5.85 to 9.29 years in the test set. The TMM method performed the best among the three models, especially in the XGBoost model; it achieved an MAE of 4.38 years in the training set and 5.85 years in the test set for the XGBoost model.

Different normalization methods have a large impact on the construction of forensic age estimation models. The TMM normalization method can effectively improve prediction accuracy and is recommended as the preferred method for forensic age estimation.