Human 5′ UTR design and variant effect prediction from a massively parallel translation assay

A fully rational approach to designing functional 5′ UTRs has not yet been implemented.

“While there has been some success in this effort in prokaryotes, yeast and even mammalian cells27,34–36, a fully rational approach to designing functional 5′ UTRs has not yet been implemented.”

• cites [27] Ferreira, J. P., Overton, K. W. & Wang, C. L. Tuning gene expression with synthetic upstream open reading frames. Proc. Natl Acad. Sci. USA 110, 11284–11289 (2013). sourcePartly supported

  The opening discussion says there were few systematic approaches for precisely controlling recombinant-protein translation in mammalian cells, which supports the claim’s general concern about design methods. It does not say that a fully rational approach to designing functional 5′ UTRs had never been implemented; the Discussion instead describes the authors’ own systematic approach using engineered RNA leaders. The claim is broader than the source supports. (LLM, conf=0.80)

  Experimental — accuracy not yet measured
• cites [34] Seo, S. W. et al. Predictive design of mRNA translation initiation region to control prokaryotic translation efficiency. Metab. Eng. 15, 67–74 (2013). sourceSupported
• cites [35] Jensen, M. K. & Keasling, J. D. Recent applications of synthetic biology tools for yeast metabolic engineering. FEMS Yeast Res. 15, 1–10 (2015). sourceCould not verify

  The editorial’s section “Yeast synthetic biology: new tools to unlock cellular function” discusses advances in synthetic-part design, but does not address 5′ UTRs or whether a fully rational approach to designing them has been implemented. Searches for 5′ UTR design and the claimed limitation found no passage that settles it; the copy has no page numbers. (LLM, conf=0.75)

  Experimental — accuracy not yet measured
• cites [36] Salis, H. M., Mirsky, E. A. & Voigt, C. A. Automated design of synthetic ribosome binding sites to control protein expression. Nat. Biotechnol. 27, 946–950 (2009). sourceDisputed

  The Materials and Methods section (Software Implementation; page unavailable) says the authors implemented an RBS Calculator that optimizes ribosome-binding-site sequences to rationally control protein expression. This contradicts the unqualified assertion that a rational design approach had not been implemented, although the paper addresses bacterial ribosome-binding sites rather than every feature of a 5′ UTR. (LLM, conf=0.77)

  Checked by a reasoning model — may be wrong.

  Experimental — accuracy not yet measured

Only 13% of human 5′ UTRs are shorter than 50 nt and thus covered by Optimus 5-Prime.

“Because only 13% of human 5′ UTRs are shorter than 50 nt and thus covered by Optimus 5-Prime, we next asked whether the approach introduced here could be extended to longer 5′ UTRs.”

• cites [44] Mignone, F. et al. Untranslated regions of mRNAs. Genome Biol. 3, reviews0004.1 (2002). sourcePartly supported

  The Structural features of untranslated regions section (page not available) says 5′ UTRs average 100–200 nucleotides and vary widely in length. It does not give the proportion of human 5′ UTRs shorter than 50 nucleotides or discuss coverage by Optimus 5-Prime. (LLM, conf=0.78)

  Experimental — accuracy not yet measured
• cites [45] Leppek, K., Das, R. & Barna, M. Functional 5′ UTR mRNA structures in eukaryotic translation regulation and how to find them. Nat. Rev. Mol. Cell Biol. 19, 158–174 (2018). sourcePartly supported

  The section ‘5′ UTRs as platforms for RNA structure’ (page unavailable) reports a median human 5′ UTR length of 218 nucleotides, supporting the general point that many are longer than 50 nt. It does not give the 13% figure or discuss which UTRs Optimus 5-Prime covers. (LLM, conf=0.84)

  Experimental — accuracy not yet measured

Earlier MPRAs designed to learn aspects of 5′ UTR cis-regulation relied on FACS or growth selection to stratify libraries by activity.

“Earlier MPRAs designed to learn aspects of 5′ UTR cis-regulation relied on FACS10,11 or growth selection12 to stratify libraries by activity.”

• cites [10] Noderer, W. L. et al. Quantitative analysis of mammalian translation initiation sites by FACS-seq. Mol. Syst. Biol. 10, 748 (2014). sourceSupported
• cites [11] Kosuri, S. et al. Composability of regulatory sequences controlling transcription and translation in Escherichia coli. Proc. Natl Acad. Sci. USA 110, 14024–14029 (2013). sourceNot checked — no full text of the source
• cites [12] Cuperus, J. T. et al. Deep learning of the regulatory grammar of yeast 5′ untranslated regions from 500,000 random sequences. Genome Res. 27, 2015–2024 (2017). sourceSupported

Translation initiation is largely dependent on start codons and their context and position relative to a CDS.

“Translation initiation is largely dependent on start codons and their context and position relative to a CDS12,17.”

• cites [12] Cuperus, J. T. et al. Deep learning of the regulatory grammar of yeast 5′ untranslated regions from 500,000 random sequences. Genome Res. 27, 2015–2024 (2017). sourceSupported
• cites [17] Hinnebusch, A. G., Ivanov, I. P. & Sonenberg, N. Translational control by 5′-untranslated regions of eukaryotic mRNAs. Science 352, 1413–1416 (2016). sourceSupported

The sequence of the 5′ untranslated region (5′ UTR) is a primary determinant of translation efficiency.

“The sequence of the 5′ untranslated region (5′ UTR) is a primary determinant of translation efficiency1,2.”

• cites [1] Araujo, P. R. et al. Before it gets started: regulating translation at the 5′ UTR. Comp. Funct. Genom. 2012, 475731 (2012). sourceSupported
• cites [2] Jackson, R. J., Hellen, C. U. T. & Pestova, T. V. The mechanism of eukaryotic translation initiation and principles of its regulation. Nat. Rev. Mol. Cell Biol. 11, 113–127 (2010). sourceSupported

The improvement in prediction accuracy from training on modified RNAs is likely due to the model learning the impact of Ψ and m1Ψ on the formation of secondary structure.

“This is likely due to the model learning the impact of Ψ and m1Ψ on the formation of secondary structure33.”

• cites [33] Kierzek, E. et al. The contribution of pseudouridine to stabilities and structure of RNAs. Nucleic Acids Res. 42, 3492–3501 (2014). sourcePartly supported

  The Results section reports that replacing U with Ψ usually increases RNA-duplex stability, supporting the premise that Ψ can affect RNA structure. The paper does not establish an effect of m1Ψ or examine whether a model learned either modification’s effects, so it cannot substantiate the proposed explanation for improved prediction accuracy. (LLM, conf=0.88)

  Experimental — accuracy not yet measured

The paper's data captured and quantitated the repressive effect of secondary structure on ribosome load.

“By calculating UTR minimum free energies (MFE)26 and comparing them to UTR MRLs, we captured and quantitated this repressive effect of secondary structure on ribosome load (Supplementary Fig. 2c)17,25.”

• cites [26] Zadeh, J. N. et al. NUPACK: analysis and design of nucleic acid systems. J. Comput. Chem. 32, 170–173 (2011). sourceNot checked — no full text of the source
• cites [17] Hinnebusch, A. G., Ivanov, I. P. & Sonenberg, N. Translational control by 5′-untranslated regions of eukaryotic mRNAs. Science 352, 1413–1416 (2016). sourceSupported
• cites [25] Kozak, M. Influences of mRNA secondary structure on initiation by eukaryotic ribosomes. Proc. Natl Acad. Sci. USA 83, 2850–2854 (1986). sourcePartly supported

  The paper reports that highly stable 5′-UTR hairpins inhibit translation by 85–95%, supporting a repressive effect of secondary structure under those conditions. It does not report quantifying ribosome load; the Discussion instead distinguishes inhibition of translation from initial 40S-subunit binding. The citing authors’ calculation of MFE and comparison with ribosome load are their own analysis, not a measurement in this source. (LLM, conf=0.84)

  Experimental — accuracy not yet measured

The model trained on the unmodified (U) library could explain 68% to 76% of the measured variability in the Ψ and m1Ψ polysome profiling data, respectively.

“We applied our method to transcripts bearing either Ψ and m1Ψ instead of uridine (U) (Fig. 2g) and found that the model trained on the unmodified (U) library could explain 68% to 76% of the measured variability in the Ψ and m1Ψ polysome profiling data, respectively (Fig. 2h).”

For the three examples, the model predicts that, of all possible variants, these specific SNVs, all of which introduce an upstream start codon, would most dramatically affect ribosome loading.

“For the three examples, the model predicts that, of all possible variants, these specific SNVs, all of which introduce an upstream start codon, would most dramatically affect ribosome loading (Fig. 4c).”

The model captured the relationship between 5′ UTR sequence and distribution of ribosome occupancy on held out test data, explaining an average of 83% of variation across all fractions.

“Using a similar network architecture but with 14 linear outputs representing the polysome fractions (Supplementary Fig. 8), the model captured the relationship between 5′ UTR sequence and distribution of ribosome occupancy on held out test data remarkably well (Fig. 2c), explaining an average of 83% of variation across all fractions (Fig. 2d).”

After polysome profiling and RNA sequencing, we retained 83,919 distinct 5′ UTRs spanning the entire length distribution from 25-100 nt.

“After polysome profiling and RNA sequencing, we retained 83,919 distinct 5′ UTRs spanning the entire length distribution from 25–100 nt.”

Sequences with out-of-frame upstream start codons (uAUGs) or upstream open reading frames (uORFs) show decreased ribosome loading.

“Our data clearly show the expected decrease in ribosome loading for sequences with either out-of-frame upstream start codons (uAUGs) (Fig. 1c) or upstream open reading frames (uORFs) (Supplementary Fig. 2b)18,19.”

• cites [18] Morris, D. R. & Geballe, A. P. Upstream open reading frames as regulators of mRNA translation. Mol. Cell. Biol. 20, 8635–8642 (2000). sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

• cites [19] Johnstone, T. G., Bazzini, A. A. & Giraldez, A. J. Upstream ORFs are prevalent translational repressors in vertebrates. EMBO J. 35, 706–723 (2016). sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

In our assay, the variant UTR of rs121908813 shows a −1.5 log2 difference in MRL compared to the wild type 5′ UTR sequence.

“In our assay, the variant UTR shows a −1.5 log2 difference in MRL compared to the wild type 5′ UTR sequence.”

Reevaluation of held-out sequences from the 'target MRL' library showed a dramatic improvement in comparison to the original model (r2 from 0.386 to 0.772).

“Reevaluation of held-out sequences from the ‘target MRL’ library showed a dramatic improvement in comparison to the original model (r2 from 0.386 to 0.772) (Fig. 3e and Supplementary Fig. 13a) as did the predicted loading of the step-wise evolved sequences (Fig. 3c red line and Supplementary Fig. 12a–d).”

Optimus 5-Prime, trained only on the eGFP library, explained 77% and 78% of the variation in MRL for two independent polysome profiling experiments of an mCherry reporter library.

“we found that the model, although only trained on the eGFP library, still performed well, explaining 77% and 78% of the variation in MRL for two independent polysome profiling experiments of this new reporter library (Supplementary Fig. 5).”

The most poorly translated sequence showed 15-fold less fluorescence than the best.

“Fluorescence and predicted MRL were highly correlated (r2: 0.87) and the most poorly translated sequence showed 15-fold less fluorescence than the best (Fig. 2e).”

CUG and GUG start codons can impact ribosome loading, especially when surrounded by strong sequence context.

“Still, we found that CUG and GUG start codons can impact ribosome loading, especially when surrounded by strong sequence context (Supplementary Figs. 3 and 4).”

Combining polysome profiling of a library of 280,000 randomized 5′ UTRs with deep learning built a predictive model that relates human 5′ UTR sequence to translation.

“Here, we combine polysome profiling of a library of 280,000 randomized 5′ UTRs with deep learning to build a predictive model that relates human 5′ UTR sequence to translation.”

• AssumptionRests on an assumption: The field lacks a means to accurately predict protein expression from 5′ UTR sequence alone.
• AssumptionRests on an assumption: The first 50 bases upstream of the CDS contain the regulatory signals that mediate translation initiation beyond ribosomal recruitment to the 5′ cap.

45 SNVs associated with human diseases substantially change ribosome loading.

“Finally, we provide evidence of 45 SNVs associated with human diseases that substantially change ribosome loading and thus may represent a molecular basis for disease.”

• AssumptionRests on an assumption: The field lacks a means to accurately predict protein expression from 5′ UTR sequence alone.
• AssumptionRests on an assumption: The first 50 bases upstream of the CDS contain the regulatory signals that mediate translation initiation beyond ribosomal recruitment to the 5′ cap.

Optimus 5-Prime MRL predictions correlated well with independently reported fluorescence levels for 77 5′ UTRs previously designed by Ferreira et al. (r2: 0.73-0.85).

“Our model’s MRL predictions correlated well with the independently reported fluorescence levels (r2: 0.73–0.85, Supplementary Fig. 9).”

• cites [27] Ferreira, J. P., Overton, K. W. & Wang, C. L. Tuning gene expression with synthetic upstream open reading frames. Proc. Natl Acad. Sci. USA 110, 11284–11289 (2013). sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

The model predicts ribosome loading of 35,000 truncated human 5′ UTRs and 3,577 naturally occurring variants.

“We test 35,000 truncated human 5′ UTRs and 3,577 naturally occurring variants and show that the model predicts ribosome loading of these sequences.”

• AssumptionRests on an assumption: The field lacks a means to accurately predict protein expression from 5′ UTR sequence alone.
• AssumptionRests on an assumption: The first 50 bases upstream of the CDS contain the regulatory signals that mediate translation initiation beyond ribosomal recruitment to the 5′ cap.

The model can explain 76% of the change of variants with log2 differences greater than 0.5 or less than −0.5.

“Importantly, the model can explain 76% of the change of variants with log2 differences greater than 0.5 or less than −0.5 (Supplementary Fig. 14a).”

Using Optimus 5-Prime and a genetic algorithm, we were able to engineer new 5′ UTR sequences for targeted ribosome loading.

“Finally, using Optimus 5-Prime and a genetic algorithm, we were able to engineer new 5′ UTR sequences for targeted ribosome loading, enabling even more forward-looking applications in synthetic biology and precision medicine.”

Optimus 5-Prime could explain 93% of MRL variation in the test set.

“After an exhaustive grid search to find optimal hyperparameters (Fig. 2a, Online Methods), Optimus 5-Prime could explain 93% of MRL variation in the test set (Fig. 2b).”

For the step-wise evolved UTRs in the second set, predicted MRLs closely matched the trend of the observed MRL along the trajectory.

“For the step-wise evolved UTRs in the second set, predicted MRLs (green) closely matched the trend of the observed (blue) along the trajectory.”

Position-specific k-mer (k = 1-6) linear models could at best explain 66% of the variation in the test set.

“By comparison, position-specific k-mer (k = 1–6) linear models could at best explain 66% of the variation in the test set (Supplementary Figs. 6 and 7, Supplementary Table 1).”

The average number of sequencing reads per library member rapidly decreased with increasing UTR length.

“In fact, we found that the average number of sequencing reads per library member rapidly decreased with increasing UTR length, likely because of the decreasing yield of full-length sequences for longer 5′ UTRs (Supplementary Fig. 15b).”

For sequences not containing uAUGs, the MRL slightly increased with length.

“Moreover, for sequences not containing uAUGs, the MRL slightly increased with length, likely because longer transcripts can accommodate more ribosomes (Supplementary Fig. 15a).”

Prediction accuracy for modified RNAs could be further improved by training the models directly on data from the modified RNAs.

“Prediction accuracy could be further improved by training the models directly on data from the modified RNAs (the same held-out library sequences were used in all test sets to ensure consistency).”

In both sets, the prediction from the model and the observed MRL eventually diverged as the model produced UTRs with very high predicted MRLs.

“While we created sequences with high ribosome loading (Supplementary Fig. 12e), in both sets the prediction from the model and the observed MRL eventually diverged as the model produced UTRs with very high predicted MRLs.”

eGFP fluorescence and predicted MRL were highly correlated (r2: 0.87).

“Fluorescence and predicted MRL were highly correlated (r2: 0.87) and the most poorly translated sequence showed 15-fold less fluorescence than the best (Fig. 2e).”

Of the 12,000 total UTRs evolved for targeted expression in the first set, the median MRL for targets 3 through 8 followed the expected trend from low to high with low variability within each group.

“Of the 12,000 total UTRs evolved for targeted expression in the first set, the median MRL for targets 3 through 8 followed the expected trend from low to high with low variability within each group.”

As observed with the 50 nt library, sequences containing uAUGs had a lower median MRL than similar length sequences not containing them.

“As observed with the 50 nt library, sequences containing uAUGs had a lower median MRL than similar length sequences not containing them.”

The same approach can be extended to chemically modified RNA.

“We show that the same approach can be extended to chemically modified RNA, an important feature for applications in mRNA therapeutics and synthetic biology.”

• AssumptionRests on an assumption: The field lacks a means to accurately predict protein expression from 5′ UTR sequence alone.
• AssumptionRests on an assumption: The first 50 bases upstream of the CDS contain the regulatory signals that mediate translation initiation beyond ribosomal recruitment to the 5′ cap.

Both sequences containing uAUGs and not containing them could span the full MRL range.

“Both sequences containing uAUGs and not containing them could span the full MRL range.”

Mean ribosome load is more positively correlated with a UTR's predicted minimum free energy (MFE) for Ψ (r = 0.56) and m1Ψ (r = 0.58) than for U (r = 0.43).

“Concordantly, mean ribosome load is more positively correlated with a UTR’s predicted minimum free energy (MFE) for Ψ (r = 0.56) and m1Ψ (r = 0.58) than for U (r = 0.43) (Fig. 2i).”

Using the retrained model, the authors were able to explain 81% of the observed variation in MRL with the common and SNV 5′ UTR sequences.

“Using the retrained model, we were able to explain 81% of the observed variation in MRL with the common and SNV 5′ UTR sequences (Fig. 4a) showing that, despite training on random sequences, the model was able to learn the cis-regulatory rules of human 5′ UTR sequences that lay directly upstream of a coding sequence.”

The analysis recapitulated the importance of a purine (A or G) at position −3 relative to AUG and a G at +4.

“This analysis recapitulated the importance of a purine (A or G) at position −3 relative to AUG and a G at +4 (Fig. 1e)10,23,24.”

• cites [10] Noderer, W. L. et al. Quantitative analysis of mammalian translation initiation sites by FACS-seq. Mol. Syst. Biol. 10, 748 (2014). sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

• cites [23] Hinnebusch, A. G. The scanning mechanism of eukaryotic translation initiation. Annu. Rev. Biochem. 83, 779–812 (2014). sourceNot checked — no full text of the source
• cites [24] Kozak, M. Point mutations define a sequence flanking the AUG initiator codon that modulates translation by eukaryotic ribosomes. Cell 44, 283–292 (1986). sourceNot checked — no full text of the source

On average, CUG and GUG are used considerably less as alternative start codons compared to AUG.

“On average, we observed considerably lower use of CUG and GUG as alternative start codons compared to AUG (Fig. 1c and Supplementary Figs. 3 and 4) unlike other reports that show widespread usage of non-AUG start sites15,20,21, possibly because these alternative start codons are used more often under stress conditions22.”

Decreased expression of TMEM127 could explain the observed pathogenicity of rs121908813.

“TMEM127 acts as a tumor suppressor and decreased expression of it could explain the observed pathogenicity of this variant.”

• AssumptionRests on an assumption: TMEM127 acts as a tumor suppressor.
• AssumptionRests on an assumption: Changes in mean ribosome load (MRL) measured in the assay reflect changes in protein expression that can affect disease pathogenicity.

The relatively lower predictive accuracy compared to direct variant effect prediction is a consequence of the increased noise due to comparing two measurements.

“The relatively lower predictive accuracy compared to direct variant effect prediction is a consequence of the increased noise due to comparing two measurements.”

UTRs with more reads have higher resolution and more accurately reflect their mean ribosome load than low-read, noisier UTRs.

“UTRs with more reads have higher resolution and so more accurately reflect their mean ribosome load compared to low read, and thus noisier UTRs (Supplementary Fig. 15a–c).”

The large MRL difference suggests that rs867711777, labeled as uncertain in the ClinVar database, could be pathogenic.

“The large MRL difference suggests that this SNV, labeled as uncertain in the ClinVar database, could be pathogenic.”

• AssumptionRests on an assumption: Changes in mean ribosome load (MRL) measured in the assay reflect changes in protein expression that can affect disease pathogenicity.

The 45 disease-associated SNVs that substantially change ribosome loading may represent a molecular basis for disease.

“Finally, we provide evidence of 45 SNVs associated with human diseases that substantially change ribosome loading and thus may represent a molecular basis for disease.”

• AssumptionRests on an assumption: The field lacks a means to accurately predict protein expression from 5′ UTR sequence alone.
• AssumptionRests on an assumption: The first 50 bases upstream of the CDS contain the regulatory signals that mediate translation initiation beyond ribosomal recruitment to the 5′ cap.

The 5′ UTR library with random sequence length 25 to 100 nt would increase the coverage of human 5′ UTRs to 29%.

“To this end, we first created a 5′ UTR library where the length of the random sequence upstream of the start codon ranged from 25 to 100 nt, which would increase the coverage of human 5′ UTRs to 29%.”

Variants within the 50 nt window directly adjacent to the start codon are under stronger negative selection than those further upstream.

“Intriguingly, variants within the 50 nt window directly adjacent to the start codon are under stronger negative selection than those further upstream16, providing another motivation to focus on this window.”

• cites [16] Whiffin, N. et al. Characterising the loss-of-function impact of 5′ untranslated region variants in whole genome sequence data from 15,708 individuals. Preprint at https://www.biorxiv.org/content/10.1101/543504v1 (2019).Not matched

  Couldn’t be matched to a paper in Crossref, so it can’t be checked.

Assessing model performance on endogenous transcripts is challenging due to confounding contributions of 3′ UTRs and coding sequence lengths.

“Assessing model performance on endogenous transcripts is challenging due to confounding contributions of 3′ UTRs and coding sequence lengths.”

A model-based approach can in principle be used to score the impact of any 5′ UTR variant on translation.

“In contrast, a model-based approach can in principle be used to score the impact of any 5′ UTR variant on translation.”

The MRL increase with length for sequences not containing uAUGs is likely because longer transcripts can accommodate more ribosomes.

“Moreover, for sequences not containing uAUGs, the MRL slightly increased with length, likely because longer transcripts can accommodate more ribosomes (Supplementary Fig. 15a).”

The data and model enabled the quantitative assessment of secondary structure, uAUGs and uORFs, Kozak sequences, and other cis-regulatory sequence elements in the context of unmodified mRNA, Ψ, and m1Ψ-modified mRNA.

“The data and model enabled the quantitative assessment of secondary structure, uAUGs and uORFs, Kozak sequences, and other cis-regulatory sequence elements in the context of unmodified mRNA, Ψ, and m1Ψ-modified mRNA.”

The decrease in accuracy of the model on the mCherry library is explained in part by differences between the eGFP and mCherry polysome profiling protocols.

“The decrease in accuracy is explained in part due to differences between the eGFP and mCherry polysome profiling protocols (Online Methods).”

The decrease in average reads with increasing UTR length is likely because of the decreasing yield of full-length sequences for longer 5′ UTRs.

“In fact, we found that the average number of sequencing reads per library member rapidly decreased with increasing UTR length, likely because of the decreasing yield of full-length sequences for longer 5′ UTRs (Supplementary Fig. 15b).”

Despite training on random sequences, the model was able to learn the cis-regulatory rules of human 5′ UTR sequences that lay directly upstream of a coding sequence.

“showing that, despite training on random sequences, the model was able to learn the cis-regulatory rules of human 5′ UTR sequences that lay directly upstream of a coding sequence.”

The slight decrease in performance observed for longer 5′ UTRs is due to lower read coverage for longer sequences and concomitant decrease in the quality of the test set.

“The slight decrease in performance observed for longer 5′ UTRs is due to lower read coverage for longer sequences and concomitant decrease in the quality of the test set.”

Each TIS sequence can uniquely tune translation initiation to a fine degree.

“Ultimately, these data suggest that each TIS sequence can uniquely tune translation initiation to a fine degree.”

FACS- and growth-selection-based MPRAs require expression of a single library variant per cell that must be transcribed within the cell from a DNA template, making it difficult to distinguish between the effects of transcriptional and translational control.

“These techniques require the expression of a single library variant per cell that must be transcribed within the cell from a DNA template, making it difficult to distinguish between the effects of transcriptional and translational control.”

• cites [10] Noderer, W. L. et al. Quantitative analysis of mammalian translation initiation sites by FACS-seq. Mol. Syst. Biol. 10, 748 (2014). sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

• cites [11] Kosuri, S. et al. Composability of regulatory sequences controlling transcription and translation in Escherichia coli. Proc. Natl Acad. Sci. USA 110, 14024–14029 (2013). sourceNot checked — no full text of the source
• cites [12] Cuperus, J. T. et al. Deep learning of the regulatory grammar of yeast 5′ untranslated regions from 500,000 random sequences. Genome Res. 27, 2015–2024 (2017). sourceNot checked yet

  Checked before — its earlier verdict is held back until a new check runs.

The approach developed here is not limited to fixed-length UTRs and could be extended even beyond a 100 nt window by synthesizing correspondingly longer 5′ UTRs for model training.

“These results suggest that the approach we developed here is not limited to fixed-length UTRs and could be extended even beyond a 100 nt window by synthesizing correspondingly longer 5′ UTRs for model training.”

Polysome profiling overcomes the limitation of single-variant-per-cell MPRAs by enabling single cells to translate tens to hundreds of in vitro transcribed (IVT) and transfected mRNA variants.

“Polysome profiling13 overcomes this limitation by enabling single cells to translate tens to hundreds of in vitro transcribed (IVT) and transfected mRNA variants.”

• cites [13] Zuccotti, P. & Modelska, A. in Post-Transcriptional Gene Regulation (ed. Dassi, E.) 59–69 (Humana Press, 2016). sourceNot checked — no full text of the source