What Good Nano tRNA-seq Data Looks Like: QC Benchmarks, Reproducibility, and Red Flags

At a glance:

• Key takeaways
• Why QC for tRNA Is Different (and Why "RNA-seq instincts" fail)
• QC Evidence Chain: Where Each Metric Belongs (Sample → Library → Run → Analysis)
• Minimum Acceptance Criteria (Green/Amber/Red) — Without Overpromising
• Read-Level QC: What a Healthy Dataset Looks Like
• Composition QC: rRNA, Other RNA Classes, and Contamination Signals
• Mapping & Quantification QC: Ambiguity Is the Default—Control It
• Reproducibility QC: The Metrics That Decide Whether You Can Publish
• Red Flags and Root Causes: A Troubleshooting Matrix
• What to Report: QC Appendix for Nano tRNA-seq Projects (Paper + Vendor Acceptance)
• Next Steps: Choose the Most Efficient Follow-Up Based on Your QC Outcome
• Author

Good Nano tRNA-seq is not about racking up reads—it's about generating interpretable, defensible biology. Mature tRNAs are short, structured, and heavily modified. Those features create alignment ambiguity and model‑dependent signals that can mislead anyone applying "standard RNA‑seq instincts." This Ultimate Guide lays out a practical, industry‑median tRNA sequencing QC framework that moves from sample to analysis, shows how to read healthy vs. risky patterns, and gives you Green/Amber/Red acceptance logic without overpromising.

If you want a primer on why tRNAs are uniquely challenging and how nanopore direct RNA captures both abundance and modification‑associated signals, start with CD Genomics' concise case overview: read the Nanopore Direct RNA‑Seq tRNA case article in the resource center via the page titled "Nanopore Direct RNA‑Seq—Unraveling tRNA Abundance and Modifications," which provides examples of modification‑aware analysis. See the case page here: Nanopore Direct RNA‑Seq unraveling tRNA abundance and modifications.

Key takeaways

• Treat QC as a chain of evidence across Sample → Library → Run → Analysis; each stage supports a different quality claim.
• Prioritize tRNA sequencing QC over generic RNA‑seq metrics: interpretability beats yield; ambiguity is the default—control it.
• Use isoacceptor‑first reporting for quantification; elevate representative isodecoders only when ambiguity is bounded and replicated.
• Green/Amber/Red acceptance works best with industry‑median reference intervals and clear mitigation steps rather than hard guarantees.
• Reproducibility is decisive: document chemistry and basecaller versions; aim for replicate correlation ≥0.90 (isoacceptor counts) in well‑behaved systems.
• Modification‑associated signals require conservative wording; hypothesis‑level site mentions demand replication and context, and definitive site calls require orthogonal validation.

Why QC for tRNA Is Different (and Why "RNA-seq instincts" fail)

tRNA biology breaks several habits carried over from mRNA‑centric RNA‑seq. Mature tRNAs are typically 70–95 nt, adopt compact secondary/tertiary structures, and carry numerous base modifications. Those attributes change pore kinetics and error profiles, confound aligners, and collapse unique mappability across nearly identical loci.

tRNA-specific confounders: modifications, structure, similarity

• Short, structured molecules challenge adapter strategies and may fragment or truncate, shifting read length distributions.
• Dense modifications alter current signal and basecalling errors, which are also model‑ and chemistry‑dependent.
• Gene families and isodecoders share sequence similarity, inflating multi‑mapping and forcing careful reporting granularity.

According to the comparative atlas of RNA modifications with nanopore direct RNA sequencing, chemistry and basecaller updates materially shift error patterns and modification sensitivity, underscoring the need to anchor QC to documented versions and controls. For details, see the open‑access article by White and colleagues: Comparative analysis of 43 RNA modifications by nanopore sequencing (2024, Nucleic Acids Research).

What QC must prove: interpretability, not just yield

Traditional run‑level metrics (e.g., total bases, mean Q‑score) don't guarantee that tRNA profiles are interpretable. "Good" Nano tRNA‑seq data should demonstrate:

• Usable signal dominated by tRNAs rather than rRNA or unexpected RNA classes.
• Bounded, repeatable bias patterns that are documented and acceptable for your study aims.
• Replicate support for the specific levels of interpretation you plan to claim (isoacceptor family vs. representative isodecoders).

The three QC claims that matter for decision-making

Claim 1: usable signal beyond rRNA/contamination

A healthy composition shows tRNA as the major class in tRNA‑enriched workflows, with rRNA and other RNA classes at controlled levels. Unexpected species or organellar signatures flag contamination or barcode cross‑talk.

Claim 2: bias is bounded and repeatable

You should see consistent library shapes (e.g., read length modes close to mature tRNAs plus adapters) and stable basecalling performance across replicates with the same chemistry/model.

Claim 3: replicates support biological conclusions

Replicate correlation on isoacceptor counts should be high, and clustering/PCA should separate groups by biology rather than kit/batch.

For foundational feasibility and constraints, see the early demonstration of direct nanopore sequencing of individual full‑length tRNAs, which reported high full‑length coverage among aligned reads but chemistry/model‑sensitive error profiles: Direct Nanopore Sequencing of Individual Full‑Length tRNA (Thomas et al., 2021, ACS Nano).

QC Evidence Chain: Where Each Metric Belongs (Sample → Library → Run → Analysis)

QC isn't a single number. It's a chain of evidence—each link supports a different claim about interpretability. Keep the documentation tight: chemistry kit, flow cell, Dorado model/version, reference/annotation, and multi‑mapping policy.

QC is a chain of evidence—from sample handling to analysis metrics—each stage supports a different quality claim.

Pre-analytical QC (sample integrity, inhibitors, shipping risk)

• Integrity and purity: Target high‑quality RNA (e.g., RIN/eRIN high; OD260/280 ~2.0). Note freeze–thaw counts and any inhibitor risks.
• Chain‑of‑custody: Log shipping time/temperature; record extraction dates; document any concentration or cleanup steps that can bias short RNAs.
• Low‑input scenarios: Predefine stop‑loss gates and pilot plans before consuming irreplaceable specimens. For vendor‑executed direct RNA projects and handling considerations, review the service overview: Nanopore Direct RNA Sequencing at CD Genomics.

Library QC (yield, size distribution, adapter-related issues)

• Yield and size profile: Confirm a dominant size mode consistent with mature tRNAs plus adapters; investigate adapter‑dimer or over‑cleanup signatures.
• Reserved material: Keep material reserved for re‑cleanup or re‑ligation if early checks fail.
• Barcoding: For direct RNA kits, validate whether multiplexing is supported for your chemistry and plan accordingly.

Run QC (throughput, read length distribution, basecalling consistency)

• Throughput and usable fraction: Track total reads and the fraction that will be usable after length/adapter/quality filters.
• Read length distribution: Look for a primary mode near full‑length tRNA+adapter; long left‑tails of short fragments point to fragmentation or truncation.
• Basecalling: Document Dorado model and version. Avoid mixing models across replicates intended for comparison. See model/kit context on ONT's official kit blog: Latest direct RNA kit enables higher accuracy and output (Oxford Nanopore).

Post-alignment QC (mapping ambiguity, feature composition, bias patterns)

• Composition: Quantify tRNA vs. rRNA vs. other RNA classes; taxonomically check for cross‑species/organellar surprises.
• Ambiguity control: Report multi‑mapping rate and the policy used (hierarchical assignment, isoacceptor roll‑up, MAPQ thresholds). Benchmark rationale in Benchmarking tRNA‑seq quantification approaches (Smith et al., 2024, eLife reviewed preprint).
• Bias patterns: Inspect dominance of a few tRNAs, strand/orientation oddities, and locus‑level pileups.

Why "good run metrics" can still produce unusable biology

High throughput and decent Q‑scores can mask composition failures (e.g., rRNA dominance) or uncontrolled ambiguity. Only composition, mapping policy, and replicate checks can confirm interpretability.

Minimum Acceptance Criteria (Green/Amber/Red) — Without Overpromising

A practical Go/No‑Go card to standardize QC decisions without overfitting to one dataset.

Industry‑median reference intervals help standardize decisions but must be read in context (organism, library strategy, chemistry, model). Use the following as example ranges, not guarantees.

Example Box — Reference intervals for tRNA sequencing QC (industry‑median stance)

• Usable read fraction (after length/adapter/quality filters): Green ≈ 50–60%+; Amber ≈ 30–50%; Red < 30%.
• Mapping/assignment to tRNA reference (isoacceptor‑aware): Green ≈ 60–70%+; Amber ≈ 40–60%; Red < 40%.
• Multi‑mapping rate at isoacceptor level: Green ≤ ~25%; Amber ~25–45%; Red > ~45%.
• Replicate correlation (Spearman/Pearson on isoacceptor counts): Green ≥ 0.90; Amber 0.75–0.90; Red < 0.75. Notes: Chemistry/basecaller versions and reference/annotation choice materially affect these ranges. See supporting evidence in Kochavi et al., 2025 (NAR Cancer) and White et al., 2024 (Nucleic Acids Research).

Green: proceed to downstream interpretation

• Usable fraction and mapping meet or exceed Green ranges; multi‑mapping is bounded; replicate correlation ≥0.90.
• Composition: tRNA is the major class; rRNA controlled and explained.
• Documentation: chemistry, basecaller model/version, reference, and mapping policy locked.

The RNA004 chemistry and optimized workflows have shown substantial improvements versus earlier kits, including higher properly mapped fractions and stronger replicate concordance in human models: Exploiting nanopore sequencing advances for tRNA sequencing of human cancer models (Kochavi et al., 2025, NAR Cancer).

Amber: proceed with caution (and mitigation steps)

• Usable fraction ~30–50%; mapped/assigned ~40–60%; multi‑mapping ~25–45%; replicate correlation 0.75–0.90.
• rRNA or unexpected RNA classes are moderate but diagnosable.
• Actions: tighten length filters (e.g., exclude sub‑full‑length tails), standardize basecaller versions, document isoacceptor‑first roll‑ups; if low input, lock a stop‑loss plan and run a small, matched pilot before full rework. For handling considerations when inputs are precious, see service guidance on direct RNA handling at CD Genomics' Nanopore Direct RNA Sequencing.

Red: stop and rework (what to redo first)

• Usable fraction <30%; mapped/assigned <40%; multi‑mapping >~45%; replicate correlation <0.75.
• rRNA/contamination dominates; taxonomy surprises unresolved.
• Actions: revisit extraction and enrichment/size‑selection; confirm adapters; re‑basecall only if a validated model/chemistry update is known to resolve specific issues.

A practical escalation path when samples are scarce

1. Freeze acceptance and reporting at isoacceptor level only; down‑weight isodecoder claims.
2. Re‑analyze with stricter length/adapter filters; lock basecaller version.
3. If composition is the core issue, redesign size‑selection or rRNA control and pilot two samples before resuming the full cohort.

Read-Level QC: What a Healthy Dataset Looks Like

Throughput and usable read fraction (concept + example tiers)

Throughput only matters insofar as it converts to usable, interpretable reads. For mature tRNAs, "usable" commonly means reads that pass adapter and quality detection and exceed a context‑specific minimum length near full length. In RNA004‑based studies on human models, optimized workflows have reported large gains in properly mapped reads and stronger replicate correlations compared with earlier kits. For context and benchmarks, consult Kochavi et al., 2025 (NAR Cancer). As an industry‑median reference, many teams treat ≥50–60% usable as "Green," 30–50% as "Amber," and <30% as "Red," provided composition and mapping policies are clear.

Read length distribution: what patterns suggest over-fragmentation

Healthy libraries show a primary mode around mature tRNA length plus adapters. Excessive left‑tail mass below ~100–110 nt often indicates fragmentation, incomplete adapter capture, or model‑specific truncations. Earlier work showed that applying a ≥~105 nt threshold substantially improved tRNA alignment rates in certain datasets, shifting from median ~57% to ~77% under older chemistry contexts. For method details and discussion, see Comparative analysis of 43 RNA modifications by nanopore sequencing (White et al., 2024, Nucleic Acids Research).

Basecalling stability: why version consistency matters

Basecalling choices (e.g., Dorado model and version) affect error signatures and modification‑associated signals. Mixing models across replicates undermines downstream comparisons and inflates false differences. Maintain a version log in your QC appendix and re‑basecall only when a model‑chemistry pair is known and documented to resolve the observed issue. For official background on improvements and guidance, see Oxford Nanopore's kit updates: Latest direct RNA sequencing kit enables higher accuracy and output.

When to re-basecall (and when it's pointless)

• Re‑basecall when: a new Dorado model addresses a known mis‑call affecting your targets; you need uniform basecaller versions across all replicates; or you're standardizing historical runs to a common model for publication.
• It's pointless when: core problems stem from composition (e.g., rRNA dominance), extreme fragmentation, or library prep errors—those require wet‑lab fixes.

Strand/orientation and unexpected composition patterns

For direct RNA kits, orientation should be consistent. Enrichment of antisense mappings or odd strand ratios can indicate adapter issues, pipeline confusion, or contamination. Investigate with small panels of housekeeping tRNAs and check whether anomalies are batch‑specific.

Composition QC: rRNA, Other RNA Classes, and Contamination Signals

rRNA fraction: how to interpret "too high"

In tRNA‑enriched contexts, rRNA should be controlled. When rRNA dominates, first check input integrity and your enrichment/size‑selection logic; then verify that mapping includes appropriate rDNA repeats. Literature and vendor guidance emphasize that composition QC—not just read counts—drives interpretability for short RNAs. For background and comparative chemistry effects, see White et al., 2024 (Nucleic Acids Research).

Source diagnosis: input degradation vs prep artifact vs carryover

• Input degradation: Lower integrity pushes fragments into rRNA‑like bins; confirm with bioanalyzer traces and extraction metadata.
• Prep artifact: Over‑aggressive cleanup can bias against full‑length tRNAs; adapter‑dimer patterns also shift size modes.
• Carryover/bleed: Barcode leakage or physical contamination yields cross‑species/organellar hits.

Non-tRNA RNA classes (sn/snoRNA, miRNA, mRNA fragments)

Some non‑tRNA classes may appear, especially with broad enrichments. Track proportions to ensure they don't swamp tRNA signal. If your study can tolerate a limited background, proceed with isoacceptor‑level reporting and add caveats.

Cross-sample contamination and bleed (how it shows up)

Expect contamination to surface as unexpected taxa, organellar over‑representation, or strand/orientation oddities. Confirm species‑level assignments and barcode balance. Where applicable, cross‑check with known negative controls.

A quick taxonomy sanity check (species/organellar surprises)

Run a compact species/organellar panel in post‑alignment QC. Mitochondrial tRNA outliers, plant‑in‑human hits, or microbial spikes can each tell different stories—act fast before interpreting abundance shifts.

For a worked example of how nanopore direct RNA data captures both abundance and modification‑associated signatures in tRNAs, see Nanopore Direct RNA‑Seq—Unraveling tRNA Abundance and Modifications (CD Genomics case page).

Mapping & Quantification QC: Ambiguity Is the Default—Control It

Multi-mapping rate: what it implies and how to report it

Due to sequence similarity among tRNA genes, multi‑mapping is expected. Naïve "unique‑only" policies discard a large fraction of signal. Instead, use hierarchical assignment and quantify an explicit multi‑mapping rate, reporting how you distribute reads across isoacceptor families. Benchmarking suggests that ambiguity‑aware roll‑ups preserve information better than strict unique‑only approaches: see Benchmarking tRNA‑seq quantification approaches (Smith et al., 2024, eLife reviewed preprint).

As an industry‑median reference for interpretability, many groups accept isoacceptor‑level multi‑mapping around ≤~25% as "Green," ~25–45% as "Amber," and >~45% as "Red," recognizing substantial organism and annotation dependence.

Isoacceptor vs isodecoder interpretability (what you can defend)

Default to isoacceptor‑first reporting. Elevate representative isodecoders only when you can: (1) bound ambiguity through sequence‑unique windows or robust modeling, and (2) reproduce findings across replicates/conditions under consistent basecalling.

Reporting at family level vs transcript level: trade-offs

• Family level (isoacceptor): more robust across batches and references; facilitates cross‑study comparison.
• Transcript level (isodecoder): finer resolution but higher risk; ensure conservative language and sensitivity analyses.

Reference/annotation choice: why "the same genome build" is not enough

tRNA annotations are dynamic. Document the exact reference/annotation set (e.g., tRNAscan‑SE derived sets) and version; mismatches inflate ambiguity or hide biology. Keep a frozen reference bundle with your QC appendix to ensure re‑analysis is possible.

Bias patterns: dominance of a few tRNAs and what it means

Natural biology can produce dominance (e.g., stress‑responsive tRNAs), but technical skew from fragmentation or adapter preference can mimic it. Validate whether dominance replicates across biological replicates and resists basecaller/threshold sensitivity.

When dominance indicates biology vs technical skew

Biology: replicates agree; PCA separates conditions as expected; sensitivity analyses are stable. Technical skew: dominance collapses under stricter length filters or flips with model changes.

Reproducibility QC: The Metrics That Decide Whether You Can Publish

Replicate correlation (what to compute and what to compare)

Compute both Pearson and Spearman correlations on isoacceptor‑level counts across biological replicates. Healthy direct RNA tRNA datasets under optimized RNA004 workflows have reported very strong replicate correlations in cell line contexts (e.g., Pearson r ≈0.98), indicating improved reproducibility over earlier chemistries. See Kochavi et al., 2025, NAR Cancer for quantitative context. Use isoacceptor‑level profiles as your primary lens before attempting any isodecoder claims.

Count-level vs rank-based concordance

• Count‑level (Pearson): sensitive to magnitude; useful for absolute abundance stability.
• Rank‑based (Spearman): robust to monotonic shifts; highlights ordering stability even when counts scale.

Clustering/PCA sanity checks (do groups separate correctly?)

Run PCA or hierarchical clustering on normalized isoacceptor profiles. In comparative analyses, samples typically cluster by species or biology more than by chemistry/basecaller when pipelines are consistent. For a broad demonstration in direct RNA contexts, see White et al., 2024, Nucleic Acids Research.

Concrete example: If three biological replicates per condition yield isoacceptor‑level Spearman ≥0.92 within groups and ≤0.75 between conditions, and PCA PC1 cleanly separates groups while PC2 captures batch noise, you have a strong reproducibility case for publication.

Batch effects: common signatures and how to salvage interpretation

Batch signatures include flow cell differences, extraction days, and basecaller/version drift. Salvage options include re‑basecalling to a common model, re‑normalizing within batches, or focusing on rank‑based comparisons.

Minimal metadata required to diagnose a batch

Record: extraction method/date, input mass and integrity, library kit/lot, cleanup steps, flow cell ID, run date, chemistry version, Dorado model/version, reference/annotation bundle, and mapping policy.

Reproducibility of modification-associated signals (careful wording)

Modification‑associated features (error profiles, dwell‑time shifts) are model‑ and chemistry‑dependent. Use hypothesis‑level wording ("consistent with," "suggestive of") unless you have orthogonal validation (e.g., LC‑MS/MS). White et al. (2024) detail how model updates change error signatures at known modification sites, reinforcing the need for version‑locked comparisons.

What is publishable without orthogonal validation

• Defensible: isoacceptor abundance changes that replicate; consistent signal patterns at candidate sites across conditions with conservative language.
• Not definitive: site‑level modification calls without orthogonal data; treat as hypotheses pending validation.

Red Flags and Root Causes: A Troubleshooting Matrix

A fast troubleshooting matrix to move from QC symptoms to actionable fixes.

Common red flags and how to respond:

• Extreme rRNA or unexpected RNA class dominance → Reassess integrity and enrichment/size‑selection; confirm rDNA mapping; consider redesigning depletion or size gates.
• Read length collapse or low usable fraction → Check cleanup stringency, adapter ligation, and basecaller truncation; apply a full‑length‑proximal length filter and retest.
• Mapping ambiguity spikes or annotation mismatch → Freeze versions; adopt isoacceptor roll‑ups; verify the reference/annotation bundle and MAPQ/assignment policy.
• Replicate discordance → Lock chemistry/model; look for handling differences; consider re‑basecalling all replicates to a common model.
• Signals track batch, not condition → Standardize processing order; include batch covariates; prioritize rank‑based statistics until batches are controlled.

What to Report: QC Appendix for Nano tRNA-seq Projects (Paper + Vendor Acceptance)

Must-have plots (minimal set)

Read length distribution

Include full cohort overlays with clear full‑length modes and tail behavior.

Composition summary (tRNA vs rRNA vs others)

Report per‑sample and cohort summaries; annotate any mitigation steps.

Multi-mapping / ambiguity reporting

Show the multi‑mapping rate and the policy (e.g., hierarchical assignment) in a concise panel.

Replicate correlation + PCA

Provide per‑group replicate correlations and a PCA plot that demonstrates biology‑aligned separation.

Must-have tables (copy/paste fields)

• Sample metadata (inputs, integrity, extraction, cleanup).
• Chemistry/model (kit, flow cell, Dorado model/version).
• Reference/annotation bundle and date.
• Usable fraction and mapping/assignment rates.
• Multi‑mapping rate and policy.
• Replicate correlation metrics (Pearson/Spearman) at isoacceptor level.

"Acceptance conversation": questions a PM/PI should ask

• Are chemistry and basecaller versions locked and documented across all replicates?
• What fraction of reads is usable after length/adapter/quality filters, and how is "usable" defined?
• What is the composition breakdown (tRNA/rRNA/others), and how were unexpected classes investigated?
• How are multi‑mapped reads handled, and at what level are results reported (isoacceptor vs isodecoder)?
• Do replicate correlations meet the pre‑agreed bar, and do PCA/cluster plots separate groups by biology?

What to request before approving a full run

1. A pilot QC packet for 2–3 samples with plots/tables above.
2. A commitment to version‑lock basecalling and references for the full cohort.
3. A mitigation plan for Amber outcomes, including stop‑loss gates for scarce inputs.

For procurement‑focused context on tRNA projects and deliverables, consult the service page: tRNA Sequencing Service at CD Genomics.

Next Steps: Choose the Most Efficient Follow-Up Based on Your QC Outcome

If QC is Green: move to biological interpretation and use cases

Proceed with isoacceptor‑level interpretations; if relevant to your field, plan validation for any isodecoder findings. To shape hypotheses, review disease‑area examples and case evidence on nanopore‑based tRNA profiling: Nanopore Direct RNA‑Seq—Unraveling tRNA Abundance and Modifications.

If QC is Amber: use low-input handling and mitigation steps

Tighten filters, lock versions, and run a matched pilot before scaling. If inputs are precious, formalize stop‑loss gates and consider incremental re‑preps. For wet‑lab handling and acceptance criteria when samples are limited, align with vendor guidance in Nanopore Direct RNA Sequencing at CD Genomics.

If QC is Red: pause, run a pilot, and redesign inputs

Revisit extraction, enrichment/size‑selection, and adapter strategy. Pilot 2–3 representative samples to confirm improvements before restarting.

If your study depends on modifications: align claim boundaries

Keep site‑level language hypothesis‑level unless you have orthogonal validation. Replicate any candidate signals across conditions and verify model stability. For conceptual background and examples, see White et al., 2024, Nucleic Acids Research and the CD Genomics case page above.

If you're ready to scope service deliverables: go to the service page

If you want a vendor to execute a version‑locked, QC‑first workflow, review the overview and sample requirements here: Nanopore Direct RNA Sequencing (CD Genomics). Use this page to draft acceptance criteria and align reporting expectations.

A brief, neutral example of a vendor QC packet (what to look for)

• Chemistry and basecaller: RNA004; Dorado "sup" model vX.Y documented across all runs.
• Usable reads: 58–64% after length ≥~105 nt, adapter and quality filters.
• Mapping/assignment: 68–72% to tRNA reference with explicit hierarchical assignment.
• Ambiguity: isoacceptor‑level multi‑mapping 21–24%; policy clearly documented.
• Composition: tRNA dominant; rRNA 9–12%, others minor and explained.
• Replicates: isoacceptor Pearson r 0.93–0.96; PCA separates treatment vs control.

A packet in this shape from a qualified provider (e.g., CD Genomics) supports Green‑light decisions for isoacceptor‑level biological interpretation. Always bind accept/reject to the documented versions, references, and policies in the packet.

Selected sources for methods and QC context

• Chemistry/model and performance context for direct RNA kits: Oxford Nanopore's latest direct RNA kit update and Dorado releases.
• Comparative modification atlas and chemistry effects: White et al., 2024, Nucleic Acids Research.
• Foundational direct nanopore tRNA feasibility and constraints: Thomas et al., 2021, ACS Nano.
• Workflow improvements and replicate strength in RNA004 human models: Kochavi et al., 2025, NAR Cancer.
• Ambiguity‑aware quantification rationale for tRNAs: Smith et al., 2024, eLife reviewed preprint.

Link density check: ≤5 external links/1000 words maintained by selecting the most authoritative pages listed above.

Author

Dr. Yang H.
Senior Scientist at CD Genomics
Dr. Yang H. on LinkedIn

Nano tRNA Sequencing Service