Journal: bioRxiv

Article Title: Genetic variation shapes human mRNA translation and disease risk

doi: 10.64898/2026.02.10.705206

Figure Lengend Snippet: ( A ) Overview of ribosome profiling and RNA-seq data processing. Transcripts per million (TPM) values were calculated from Ribo-seq and RNA-seq data, and translation efficiency (TE) was calculated as TPM ribo-seq /TPM RNA-seq . mRNAs were classified into high-, intermediate-, and low-TE categories, with “individual-sharing” transcripts defined by consistent classification across samples after filtering (see Methods). ( B ) Architecture of the convolutional-recurrent hybrid neural network for TE prediction. Full-length mRNA sequences were one-hot encoded, with a fifth channel labeling RNA regions (5’UTR, CDS, and 3’UTR), and fed into the CNN and BiLSTM layers, followed by fully connected layers to output the predicted TE-high probabilities (referred to as pTE). ( C ) Model performance metrics (AUC, PRAUC, Precision, Recall, Accuracy, and F1-score) from 10-fold cross-validation; each point represents one-fold. ( D ) Receiver operating characteristic (ROC) curves for the 10 folds; the x-axis shows specificity; the y-axis shows sensitivity. ( E ) Relationship between predicted pTE categories and actual TE values. The x-axis shows predicted score categories; the y-axis shows log 10 (TE+1) across samples to accommodate zero values.

Article Snippet: To evaluate the impact of 5’UTR variants on translation efficiency, we synthesized variant-containing 5’UTR sequences (GENEWIZ) and cloned them into the Fluc-Nluc dual luciferase reporter vector ( ).

Techniques: RNA Sequencing, Labeling, Biomarker Discovery

Journal: bioRxiv

Article Title: Genetic variation shapes human mRNA translation and disease risk

doi: 10.64898/2026.02.10.705206

Figure Lengend Snippet: Sequence and structural correlates of translation efficiency captured by TEFL-mRNA. ( A-B ) Distribution of pTE across mRNAs with or without (A) 3’UTR miRNA binding sites or (B) 5’UTR upstream open reading frame (uORF). Y-axis shows the fraction of mRNAs in each pTE category. P-value from Fisher’s Exact test indicates significant shifts in pTE distributions. ( C ) Correlation between estimated mRNA half-life and predicted TE. ( D ) Correlations between pTE and the length of the 5’UTR, CDS, and 3’UTR. ( E-G ) Correlations between pTE and (E) GC ratio, (F) codon adaptation index (CAI), or (G) adjusted minimum free energy (AMFE; minimum free energy normalized by mRNA length). For all scatter plots, points represent bin-averaged (10 mRNAs per bin) values of sequence features (x-axis) and pTE (y-axis); Pearson’s correlation coefficients (R) and p values are shown.

Article Snippet: To evaluate the impact of 5’UTR variants on translation efficiency, we synthesized variant-containing 5’UTR sequences (GENEWIZ) and cloned them into the Fluc-Nluc dual luciferase reporter vector ( ).

Techniques: Sequencing, Binding Assay

Journal: bioRxiv

Article Title: Genetic variation shapes human mRNA translation and disease risk

doi: 10.64898/2026.02.10.705206

Figure Lengend Snippet: The effects of single-nucleotide variants on translation efficiency. ( A ) Overview of TEFL-mRNA variant analysis. Genetic variants from gnomAD were annotated with VEP, and mRNA transcripts containing 5’UTR, 3’UTR, or CDS were extracted. Full-length mRNA sequences with reference and alternative alleles were input into TEFL-mRNA to calculate TE changes (ΔpTE = pTE ALT – pTE REF ). ( B ) Distribution of pTE categories for mRNAs with uAUG-creating SNVs versus all gnomAD 5’UTR variants; enrichment tested by Fisher’s Exact test. X-axis shows the fraction of mRNAs in each pTE category. ( C ) Cumulative distribution of pTE for uAUG-creating (red line) versus gnomAD 5’UTR (grey line) variants; p-value from Kolmogorov-Smirnov test. ( D ) Regional distribution of TE effects for TE-altering variants. Left: boxplot of |ΔpTE| of variants with |ΔpTE| > 0.1 in different regions. The number of such variants is shown above each box. Right: fraction of SNVs passing different |ΔpTE| cutoffs. ( E ) Relationship between allele frequency (AF) and |ΔpTE| for 5’UTR SNVs. Brown points mark variants with |pTE| > 0.1. ( F ) Percentage of transcripts with 5’UTR SNVs passing the specified |ΔpTE| cutoff within the allele frequency (AF) range among all transcripts with 5’UTR SNVs that pass the |ΔpTE| cutoff. ( G ) Schematic of the Fluc-Nluc dual luciferase reporter used to assay SNV effects on TE. ( H ) Reporter assay results for six gnomAD SNVs. Left: SNVs whose variant alleles increased TE; right: SNVs whose variant alleles decreased TE. Bars show mean ± standard deviation (SD) from biological triplicates. Y-axis shows the normalized luciferase ratio. P values from one-tailed t-tests are shown. TEFL-mRNA predicted pTE values are shown below each panel.

Article Snippet: To evaluate the impact of 5’UTR variants on translation efficiency, we synthesized variant-containing 5’UTR sequences (GENEWIZ) and cloned them into the Fluc-Nluc dual luciferase reporter vector ( ).

Techniques: Variant Assay, Luciferase, Reporter Assay, Standard Deviation, One-tailed Test

Journal: bioRxiv

Article Title: Genetic variation shapes human mRNA translation and disease risk

doi: 10.64898/2026.02.10.705206

Figure Lengend Snippet: TE-altering SNVs in coding and noncoding regions. ( A ) Predicted ΔpTE for SNVs located within ±1000nt of the start or stop codon. Each point indicates an individual SNV; colors indicate the alternative allele. X-axis shows the relative distance to the start or stop codon. ( B ) Fractions of transcripts with 4 alternative allele types (A/C/G/U) among 5’UTR SNVs exceeding different ΔpTE cutoffs. ( C ) Counts of TE-increasing and TE-decreasing synonymous SNVs with |ΔpTE| > 0.05 (corresponding to a false positive rate of 1%, similarly in panels D-G below). Variants changing codons toward non-optimal (orange) or optimal (yellow) are shown separately. P-value from Fisher’s exact test. ( D-E ) ΔpTE for missense SNVs introducing specific amino acids. (D) TE-increasing variants frequently introduce hydrophobic residues (Phe, Leu). (E) TE-decreasing variants more often introduce Gly, Thr, and Pro, linked to reduced elongation efficiency. ( F ) Proportion of TE-increasing versus TE-decreasing effects among all missense SNVs or Pro-substituting SNVs; enrichment tested by Fisher’s Exact test. Y-axis shows the fraction of transcripts with the SNVs in each category. ( G ) Relationship between ΔpTE and poly-proline tract length created by missense SNVs. n denotes the number of SNVs per group.

Article Snippet: To evaluate the impact of 5’UTR variants on translation efficiency, we synthesized variant-containing 5’UTR sequences (GENEWIZ) and cloned them into the Fluc-Nluc dual luciferase reporter vector ( ).

Techniques: Introduce

Journal: bioRxiv

Article Title: Genetic variation shapes human mRNA translation and disease risk

doi: 10.64898/2026.02.10.705206

Figure Lengend Snippet: Contribution of mRNA structure and RBP binding to translation alteration. ( A ) Absolute changes in adjusted minimum free energy (|ΔAMFE|) for TE-decreasing, TE-increasing, and TE-unchanged SNVs in the 5’UTR and CDS. ( B ) Comparison of |ΔAMFE| between TE-decreasing and TE-increasing SNVs in the 5’UTR and CDS; p-values from Student’s t-test are shown. ( C ) Structures of the mRNA NM_001320334 with the reference (C, left) or alternative allele (T, right) at an SNV (rs758379451) in the 5’ UTR. Structures were predicted using LinearFold and visualized with foRNA. ( D ) Heatmaps showing RBPs whose binding was significantly changed by TE-altering SNVs (compared to TE-unchanged SNVs) in 5’UTR (left) and CDS (right). Binding changes were predicted by DeepBind; p-values from Student’s t-test are color-coded. ( E-F ) Representative RBP binding motifs in the 5’UTR (E) and CDS (F). For each motif, bar plots show counts of TE-altering SNVs that map to specific motif positions.

Article Snippet: To evaluate the impact of 5’UTR variants on translation efficiency, we synthesized variant-containing 5’UTR sequences (GENEWIZ) and cloned them into the Fluc-Nluc dual luciferase reporter vector ( ).

Techniques: Binding Assay, Comparison

Journal: bioRxiv

Article Title: Genetic variation shapes human mRNA translation and disease risk

doi: 10.64898/2026.02.10.705206

Figure Lengend Snippet: Disease associations of genetic variants that alter translation. ( A ) Construction and evaluation of processes for variants with |ΔpTE| > 0.1. Dot size indicates the number of genes the GO term contains; colors denote parent terms summarized by rrvgo . Axes represent the first two components from PCoA of the semantic similarity matrix. ( B ) Disease ontology enrichment for the same variant set as in (A). Dot size indicates gene counts; color scale shows adjusted p-values. ( C ) Fractions of transcripts with 5’UTR SNVs that passed various |ΔpTE| cutoffs in GWAS (red) and gnomAD (blue). P-values from Fisher’s exact test. ( D ) GWAS diseases (left) and non-disease traits (right) enriched for 5’UTR TE-altering SNVs (|ΔpTE| > 0.1). Bars show counts; color scale indicates the fraction of transcripts in each trait carrying such variants. The enriched traits with top 10 SNVs counts were filtered using p < 0.05 from Fisher’s Exact test. ( E ) Top 10 ClinVar immune-related SNVs ranked by |ΔpTE| in the 5’UTR (top) and CDS (bottom). ( F-G ) Luciferase reporter assays for GWAS disease-related SNVs. Bars show mean ± SD from biological triplicates; points indicate replicates; y-axis shows the normalized luciferase ratio. P-values from one-tailed t-tests are shown. TEFL-mRNA predicted pTE values for reference (REF) and alternative (ALT) alleles, along with associated GWAS traits, are listed below each plot.

Article Snippet: To evaluate the impact of 5’UTR variants on translation efficiency, we synthesized variant-containing 5’UTR sequences (GENEWIZ) and cloned them into the Fluc-Nluc dual luciferase reporter vector ( ).

Techniques: Variant Assay, Luciferase, One-tailed Test

Journal: bioRxiv

Article Title: Genetic variation shapes human mRNA translation and disease risk

doi: 10.64898/2026.02.10.705206

Figure Lengend Snippet: Translation-altering SNVs across multiple cell types. ( A ) Receiver-operating characteristic (ROC) curves for TEFL-mRNA models trained separately in nine human cell types. Performance was assessed by 10-fold cross-validation for the TE-high vs others classification; each fold is shown, and the per-model AUC is annotated. ( B ) Cross-cell-type agreement for 5’UTR SNVs. Tiles show the Spearman correlations of ΔpTE between each pair of cell types, computed on ExAC 5’UTR SNVs with |ΔpTE|>0.02 in both cell types (two-sided test; *** p < 0.001). Overlaid pie charts indicate the proportion of shared SNVs with concordant positive (red), concordant negative (blue), or discordant (salmon) signs of TE effects. ( C ) Sharing of TE effects across cell lines by region. Stacked bars give the percentage of SNVs that are “shared” across cell types (orange), defined by a concordance score >0.8 (the fraction of profiled cell types in which the ΔpTE sign matches the majority sign among cell types where |ΔpTE|>0.02), for 5’UTR, CDS missense, CDS synonymous, and 3’UTR SNVs. Count and percentage of SNVs contributing to each bar are shown. ( D ) Disease enrichment of TE-altering GWAS 5’UTR SNVs across cell types. Heatmap shows the enrichment of TE-altering SNVs (|ΔpTE|>0.02) associated with disease traits (EFO:0000408) in at least two cell types using Fisher’s exact test (p < 0.05); values are z-scores of the odds ratio in each trait across cell types. Row/column dendrograms indicate hierarchical clustering of traits and cell types (*** p<0.001, ** p<0.01, * p<0.05).

Article Snippet: To evaluate the impact of 5’UTR variants on translation efficiency, we synthesized variant-containing 5’UTR sequences (GENEWIZ) and cloned them into the Fluc-Nluc dual luciferase reporter vector ( ).

Techniques: Biomarker Discovery