# UniProt Expression and Localization Data for Non-Model Organisms UniProt does not measure expression itself. It stores curated, evidence-tagged annotations about [protein function](/blog/guides/protein-function), subcellular localization, and tissue or developmental specificity, and it links out to resources that hold measured expression values. For a non-model organism the rule of thumb is simple: start from your species' proteome in UniProtKB, treat unreviewed entries as computationally derived hypotheses rather than facts, pull measured expression from connected resources such as Expression Atlas and MOPED where they cover your species, and plan to validate anything you intend to publish with your own experiments or primary literature. This guide walks through what uniprot non-model organism expression data actually contains, the main options for obtaining it, the pros and cons of each, a stepwise uniprot database tutorial you can run today, and the mistakes that most often derail non-model organism projects. ## What UniProt Actually Stores for Expression and Localization The UniProt Knowledgebase (UniProtKB) is a collection of sequences and annotations for over 120 million proteins across all branches of life. Detailed annotations extracted from the literature by expert curators cover more than half a million of these proteins, and the remainder are supplemented by rule-based automated annotation systems and imports from other resources [1]. Understanding that split is the foundation of every decision you will make in this workflow. Three UniProt concepts matter most for expression and localization work: **Reviewed versus unreviewed entries.** UniProtKB contains two sections. Reviewed entries (historically called Swiss-Prot) carry manual annotations produced by expert curators who read the underlying literature. Unreviewed entries (TrEMBL) carry annotations generated by automated systems, such as rule-based methods that transfer annotation from reviewed proteins to similar sequences. For a non-model organism, your protein of interest often lives only in TrEMBL, which means its localization statement may be a computational inference rather than an observation. **Evidence and protein existence tags.** Every annotation in a UniProt entry carries an evidence tag describing how the claim is supported, from direct experimental evidence through to computational or electronic annotation with no experimental basis. Entries also carry a Protein Existence (PE) level from 1 to 5, where level 1 indicates protein-level experimental evidence and level 5 indicates uncertainty about the protein's existence at all. In practice, a TrEMBL entry for an orphan beetle gene with PE level 4 (predicted) and a localization copied from a fly ortholog is a hypothesis about subcellular localization, not a measurement. **Structured subcellular location terms.** Subcellular location is stored in a controlled vocabulary with defined compartments such as nucleus, cytosol, mitochondrion, plasma membrane, and extracellular space, with qualifiers that distinguish fully annotated locations from by-similarity or probable locations. The UniProt website provides dedicated visualizations for the subcellular localization of proteins, alongside structure and interaction views, so a single entry page can communicate both the compartment and the confidence behind it [1]. Expression-related annotation in UniProtKB includes free-text fields such as tissue specificity, developmental stage, and induction, which curators fill from experimental papers. These fields are informative but not quantitative. They tell you that a protein was reported in liver, not how abundant it was or under what conditions. ## The Main Options for Non-Model Organism Expression and Localization Data Most projects combine several sources. The table below is the decision core of this article, and the sections after it explain each option on its own terms. | Option | What it provides | Non-model coverage | Evidence strength | Best for | Main limitation | |--|--|--|--|--|--| | Reviewed UniProtKB entries | Manually curated function, localization, tissue specificity with evidence tags | Limited to species with curated literature | Highest, backed by cited experiments | Authoritative claims for genes with direct homolog studies | Only a small fraction of species have curated entries | | Unreviewed UniProtKB (TrEMBL) entries | Rule-based automated annotation transferred from related proteins | Very broad, effectively all sequenced organisms | Variable, often computational inference only | Genome-scale first-pass annotation | Easy to mistake computed claims for measurements | | Species proteomes and ortholog transfer | Complete predicted protein sets per species, plus cross-species annotation via orthologs | Any organism with a genome or reference proteome | Depends on orthology confidence and phylogenetic distance | Annotation of newly sequenced species | Divergent lineages accumulate transfer errors | | Expression Atlas | Curated baseline and differential gene and protein expression across tissues, cell types, and conditions, integrated into UniProt | Grows with community submissions, currently spanning dozens of species | High for included studies, absent elsewhere | Asking whether an ortholog is expressed in a comparable tissue or cell type | Coverage concentrates on well-studied species | | MOPED | Protein-level expression from mass spectrometry studies in humans and model organisms, browsable by organism, tissue, localization, and condition, with links to UniProt | Humans and established model organisms | Quantified spectra with false discovery rate reporting | Checking whether a conserved protein was detected proteomically in a matching tissue | Does not cover most non-model species directly | | Primary literature and data journals | Immunofluorescence, western blot, fractionation, and imaging data for specific proteins | Any species reported in the literature | As strong as the underlying methods | Validating a specific candidate before experiments | Slow to retrieve, heterogeneous formats | | De novo experiments in your organism | Endogenous tagging, immunostaining, proteomics, single-cell protein profiling | Unlimited, you generate it | Definitive for your condition when well controlled | Publication-grade localization and expression claims | Cost and time, especially for antibody or tag generation | ### Reviewed UniProtKB Entries A reviewed entry is the gold standard of uniprot expression and localization data for non-model organisms when one exists for your protein or a very close relative. Curators extract expression and localization statements directly from papers, so a reviewed entry stating "nucleus" with an experimental evidence tag reflects a published observation, and the citation is embedded in the entry. For species adjacent to well-studied clades, for example livestock species or emerging aquaculture fish, a surprising number of conserved proteins already have reviewed homolog entries you can borrow from while planning experiments. ### Unreviewed TrEMBL Entries TrEMBL is where most non-model organism sequences live. Annotations arrive through automated systems, most importantly rule-based methods that recognize protein families, domains, and signature motifs, then propagate standardized function and localization statements [1]. This is the reason TrEMBL delivers coverage at a scale manual curation cannot match, and it is also the reason you must read evidence tags before you trust a compartment assignment. A "mitochondrion" label on an unreviewed entry may come from a transit peptide prediction, a domain signature, or a transfer from a reviewed ortholog. Each of those carries different weight. ### Species Proteomes and Ortholog-Based Annotation Every organism with a genome assembly can have a reference proteome in UniProt, meaning a curated set of protein sequences for the species. For species without one, you can still work at the level of translated transcripts from your own assembly. The practical bridge to functional annotation is orthology. Orthologs are genes in different species that descended from a common ancestral gene, and they are the primary vehicle by which annotation moves from model organisms to everything else. This bridging is an active research problem in its own right. The Multiple model species selection (MSS) approach tackles a specific failure mode: using a single reference species for annotating a non-model transcriptome loses information that closer relatives hold. MSS selects effective reference model species through taxonomic associations and ultra-conserved orthologous gene comparisons, drawing comprehensive genomic annotations from 291 reference eukaryotic species retrieved from RefSeq, KEGG, and UniProt databases, and achieved roughly a twenty-fold reduction in computational time with more accurate results than single-reference approaches [2]. The takeaway for your workflow is that the best uniprot non-model organism expression strategy rarely uses a single model organism. It uses the closest available relatives. That principle extends to literature mining. Work on bovine embryo development showed that knowledge about ruminant [gene regulation](/blog/guides/gene-regulation) is best reconstructed by considering closely related species such as sheep and pig alongside deep human and mouse literature, precisely because model organism processes have features that differ from those in the species under study, and that computational approaches integrating model organism knowledge while respecting species specificity are needed when manual processing of the literature is impractical [3]. ### Expression Atlas Expression Atlas is EMBL-EBI's knowledgebase for gene and protein expression across tissues, cell types, and conditions and across multiple species. It has grown to include more than 4,500 studies from 67 species, with increased proteomics coverage, updated GTEx tissue profiles, and hundreds of single-cell RNA-seq experiments spanning 21 species. Its data are integrated directly into EMBL-EBI resources including Ensembl, UniProt, and Europe PMC, and it is disseminated through collaborations with model organism communities such as FlyBase and Gramene [4]. For uniprot non-model organism expression questions, this integration matters because it means the UniProt entry you are reading can be a doorway to measured expression profiles, not just curator text. Coverage is the honest caveat. Species representation concentrates where data generation has concentrated. If your organism is a livestock species, a crop, or a widely studied fish, check Expression Atlas before assuming you have no expression data. If your organism is a newly described invertebrate, expect to rely on orthologs and your own experiments instead. ### MOPED The Model Organism Protein Expression Database (MOPED) enables rapid browsing of protein expression information from publicly available mass spectrometry studies on humans and model organisms. It provides protein-level expression data, meta-analysis capabilities, and quantitative data from standardized analysis, and data can be queried by protein or browsed by organism, tissue, localization, and condition, sorted by false discovery rate and expression. MOPED contains over 43,000 proteins with at least one spectral match and more than 11 million high-certainty spectra, and it links outward to protein and pathway databases including UniProt, GeneCards, Entrez, KEGG, and Reactome [5]. MOPED's role in a non-model workflow is indirect but useful. When your candidate gene has a one-to-one ortholog in human or a classic model organism, MOPED tells you whether that ortholog was detected proteomically in a tissue matching your question, and with what quantitative support. That is often the strongest pre-experimental evidence available for an expression claim. ### Primary Literature and Data Journals Many expression and localization datasets that never make it into a database live in data journals. Typical examples include verification datasets for a single protein family, such as data on a transporter's expression in cancer cell lines measured by RT-PCR and western blot with subcellular localization assessed by immunofluorescence colocalization and cellular fractionation, cross-referenced against [Human Protein Atlas](/blog/guides/human-protein-atlas) tissue microarrays [6]. Similar datasets report invadopodia-associated proteins in pancreatic cancer cells and tumors using immunofluorescence, western blotting, and co-immunofluorescence across cell lines and surgical specimens [7]. When you are validating a specific candidate before committing to experiments, searching for exactly these kinds of companion datasets for the ortholog is one of the highest-yield hours you can spend. ### De Novo Experiments in Your Organism Ultimately, the strongest uniprot non-model organism expression option is to generate the data yourself and contribute it back. The experimental logic is well established. Antibody-based detection of every protein in a complex organism is not feasible given cost, time, and labor constraints, and antibodies also cannot support live imaging of protein dynamics. That is why standardized immunoepitope or fluorescent tags, inserted at the endogenous locus or within large genomic clones, have become the preferred approach, because endogenous-locus tagging often reports proper expression, subcellular localization, and dynamics, while enabling protein removal strategies, standardized visualization, and interaction studies by mass spectrometry. Drosophila is the case study in the literature, but the logic transfers to any organism with [genome editing](/blog/guides/genome-editing) [8]. For tissue-level work, semi-quantitative immunohistochemistry with software-based analysis is a practical middle ground. An integrated protocol using ImageJ Fiji demonstrates how to deconvolve hematoxylin and DAB-stained images and quantify expression, providing stepwise reproducibility that many published methods sections lack [9]. ## Comparing the Options on the Criteria That Decide Real Choices The uniprot non-model organism expression pros and cons reduce to five criteria. **Evidence quality.** Reviewed entries beat TrEMBL, measured data beats prediction, and your own experiment beats everything. The trap is that UniProt entries of all types look visually identical on the website. Train yourself to check the evidence tag on every localization claim before it enters a manuscript. **Coverage.** TrEMBL and reference proteomes cover essentially anything sequenced. Expression Atlas covers dozens of species with thousands of studies [4]. MOPED covers humans and classic model organisms [5]. Curated entries cover the well-mined slice of biology. Your species sits somewhere on that gradient, and the option you lean on should match your position on it. **Quantitation.** UniProt localization annotation is categorical. MOPED offers quantitative proteomics with false discovery rate sorting [5]. Expression Atlas offers differential and baseline expression with statistical context [4]. If your question requires abundance, the databases built for measured data are the right tools and UniProt is the index into them. **Species fidelity.** Annotation transferred from an ortholog describes the ortholog's biology until proven otherwise. The MSS result that multiple reference species outperform single references, with large time savings, formalizes the intuition that annotation quality improves as you move down the taxonomy toward your organism [2]. **Cost of verification.** Databases are free to query but their claims arrive as given. Experiments are expensive but produce claims you own. A balanced project uses databases to eliminate dead ends cheaply and reserves experiments for the handful of candidates that matter. ## How to Use UniProt for a Non-Model Organism: A Stepwise Tutorial This is the operational heart of the guide. Each step uses only official UniProt interfaces and standard bioinformatics practice. ### Step 1: Find Your Species' Proteome Go to uniprot.org and use the proteome search, or query the REST API. For example, for cattle (NCBI taxon identifier 9913): ```bash curl "https://rest.uniprot.org/proteomes/search?query=taxon_id:9913&format=tsv&fields=upid,organism_name,protein_count" ``` If a reference proteome exists, download it and use it as the target for mapping your transcripts or proteins. If not, plan to work from your own assembly and rely on orthology more heavily. ### Step 2: Read an Entry Like a Curator ![Decision path that checks evidence tag, protein existence level and ortholog source before a localization claim is used.](/article-images/illustrations/uniprot-non-model-organism-expression-evidence-triage.webp) Open any UniProt entry and go to the "Subcellular location" section. Three things to read in order: 1. The location terms themselves, including qualifiers such as "By similarity" or "Probable." 2. The evidence tags attached to each claim, which distinguish experimental evidence from computational inference. 3. The Protein Existence (PE) level near the top of the entry. If the location is "By similarity," trace the cited source entry. That chain tells you exactly which organism's data your annotation ultimately rests on. ### Step 3: Query the Website for Localization-Specific Sets The UniProt website supports fielded queries. To pull all reviewed cattle proteins annotated to the plasma membrane, use a query such as: ``` (organism_id:9913) AND (reviewed:true) AND (subcellular_location:plasma membrane) ``` To find tissue-specificity statements, search within the entry text for tissue names, or download full entries and parse the tissue specificity comments programmatically. ### Step 4: Use Programmatic Access for Scale The UniProt REST API returns tab-separated fields suitable for downstream pipelines. To fetch accession, name, subcellular location, and tissue specificity for an entire species: ```bash curl "https://rest.uniprot.org/uniprotkb/stream?query=organism_id:9913&format=tsv&fields=accession,id,protein_name,cc_subcellular_location,cc_tissue_specificity" > cattle_uniprot_annotations.tsv ``` For an ID-mapped list from your own transcriptome, upload your gene identifiers through the ID mapping service and retrieve the corresponding UniProt accessions, then attach the annotation table by accession. This two-step pattern, map then annotate, is the backbone of nearly every uniprot non-model organism expression pipeline. ### Step 5: Bridge to Model Organisms Through Orthologs Use UniProt's orthology views or a dedicated orthology tool to identify one-to-one orthologs in human, mouse, rat, zebrafish, fly, or yeast. Then pull expression evidence for the orthologs from Expression Atlas, which you can reach from the UniProt entry through its integrated EMBL-EBI links [4], and from MOPED for proteomic detection records in matching tissues [5]. Follow the MSS principle and consult the closest relatives rather than defaulting to mouse [2]. ### Step 6: Chase the Primary Evidence For every claim you intend to build a hypothesis on, open the cited references inside the UniProt entry. Curators do the extraction, but the depth of the original experiment, for example whether localization was shown by fractionation, immunofluorescence, or fused-protein overexpression, is only visible in the source paper. Companion data articles reporting a protein's expression and localization across cell lines and tissues with defined methods are frequently available for well-studied proteins and are worth searching for directly [6], [7]. ## Worked Scenarios ### Scenario 1: Annotating a De Novo Transcriptome for an Orphan Species ![Diagram of steps from transcript sequences to a multi-species orthology assignment, bulk UniProt retrieval and an evidence-tagged table.](/article-images/illustrations/uniprot-non-model-organism-expression-annotation-workflow.webp) You have assembled a transcriptome for a non-model mollusk and want expression and localization context for 2,000 differentially expressed transcripts. The efficient path is: translate the transcripts, run orthology assignment against several model proteomes rather than one, following the multi-species logic shown to outperform single references [2], retrieve UniProt annotations for the matched accessions in bulk via the REST API, pull Expression Atlas tissue profiles for the closest relative available [4], and record for each transcript whether the localization statement is experimental in the source ortholog or merely transferred. The output is an annotation table where every localization claim carries its evidence provenance, which is what reviewers will ask for. ### Scenario 2: Prioritizing Candidate Genes Using Localization and Networks Suppose your goal is candidate gene discovery in a disease context using a non-model organism with a well-annotated relative. Subcellular compartment information gains power when integrated with protein-protein interaction networks. In one published approach, integrating subcellular localization data with PPINs, defining compartment significance, building edge-weighted networks, and detecting protein complexes allowed prioritization of novel breast cancer candidate genes, with the authors noting the technique applies to other disorders [10]. For a non-model project, the workflow translates directly: map your proteins to the better-annotated relative, weight network edges using shared localization, and prioritize genes whose compartment assignment is independently supported rather than predicted alone. ### Scenario 3: Validating a Localization Claim Before Committing to Experiments You plan to study a signaling protein in a non-model fish and a TrEMBL entry says "plasma membrane." Before ordering reagents, check the evidence tag, inspect the ortholog's reviewed entry in zebrafish or human, search Expression Atlas for tissue-level expression of the ortholog [4], and search the literature for companion localization datasets with methods such as immunofluorescence colocalization and cellular fractionation [6]. If the trail shows the compartment claim rests on a single overexpression study in a distant species, budget for endogenous validation. Endogenous-locus tagging reports localization and dynamics that overexpression constructs and cross-species antibodies often cannot [8], and if antibody work is your fallback, a documented semi-quantitative IHC pipeline in ImageJ Fiji gives you reproducible quantification rather than subjective scoring [9]. ## Localization Data Is More Context-Dependent Than Most Pipelines Assume Two findings from the primary literature deserve their own section because they change how you should interpret any localization annotation, including UniProt's. First, localization can depend on transcript processing, not just [protein sequence](/blog/guides/protein-sequence). The LABRAT study of alternative polyadenylation found consistent relationships between the use of gene-distal polyadenylation sites and subcellular [RNA localization](/knowledge/molecular-biology/rna-localization) across multiple cell types, and implicated dozens of RNA-binding proteins in driving directional shifts in alternative polyadenylation isoform abundance [11]. A protein whose mRNA localizes differently between isoforms may show cell-type-specific patterns that no single database entry captures. Second, localization depends on activation state and tissue context. A cartographic study of the aryl hydrocarbon receptor in the human placenta found constitutive nuclear localization in every placental model examined, in the absence of deliberately added exogenous activators, consistent with intrinsic activation by endogenous placental ligands [12]. The general lesson transfers: an entry stating "cytoplasm, nucleus" may reflect ligand-dependent shuttling that your static annotation flattens. And for quantitative protein measurements in single cells, technical noise from sources such as unbound antibodies captured during droplet generation requires explicit correction methods like dsb, which estimates and removes that background using empty droplets and shared variance structure [13]. Measured localization and expression data carry their own analysis burden. ## Common Mistakes and Limitations **Treating unreviewed annotations as observations.** The most common failure in non-model work. A TrEMBL localization statement generated by rule-based automated systems is a computational inference [1]. Check evidence tags and PE levels on every entry you cite. **Defaulting to a single model organism.** Evidence from the MSS study shows that selecting multiple reference species through taxonomic relationships and ultra-conserved orthologs yields more accurate annotation than single-reference approaches, with roughly a twenty-fold reduction in computation time as a bonus [2]. Locking your pipeline to mouse or human discards signal from closer relatives. **Ignoring gene dictionary size in close relatives.** Literature mining work on ruminant embryo development highlighted that gene dictionaries in some close relatives are smaller than in the species of interest, creating a real risk of reducing the resource base available for mining, and that species-specific consideration of related organisms is necessary rather than optional [3]. When your annotation bridge depends on a species with sparse identifiers, map identifiers carefully and record losses. **Reading static localization as universal.** Localization annotation compresses condition-dependent behavior into one line. Transcript isoform choice alters RNA localization [11], activation state alters protein compartmentalization [12], and developmental stage alters both expression and localization patterns in ways documented even in classic single-protein studies. Interpret entries as summaries over conditions, not descriptions of your condition. **Over-trusting antibody-based localization.** Cross-species antibodies frequently cross-react, and generating specific antibodies against all proteins of a complex organism is not feasible, which is exactly why endogenous tagging approaches were developed and shown to report proper expression, subcellular localization, and dynamics more faithfully [8]. If a literature localization claim rests on an antibody in a distant species, weight it accordingly. **Skipping quantification of imaging data.** Studies that quantify immunohistochemistry without documented image analysis are hard to reproduce or compare. Where you need semi-quantitative tissue-level numbers, follow an explicit deconvolution and measurement protocol rather than categorical scoring [9]. **Assuming single-cell protein data are as clean as single-cell RNA data.** Droplet-based protein profiling with oligo-conjugated antibodies carries substantial technical noise from unbound antibody encapsulation and cell-to-cell background variation, which motivated dedicated denoising methods validated across multiple technologies including CITE-seq and TEA-seq [13]. If you integrate such data, normalize with methods that model these specific noise sources. **Forgetting that UniProt is an index, not a warehouse of measurements.** UniProt's value is in curated, evidence-tagged summaries and its links to measured data in Expression Atlas and elsewhere, [4]. Projects fail when they expect UniProt to answer quantitative expression questions directly. Route those questions to the measurement resources and return to UniProt for the annotation frame. **One overarching limitation.** Databases describe populations of published observations, not your specific sample, organism, or condition. Individual experimental decisions, especially anything intended for publication or application, need validation in your own system by someone with hands-on expertise in that organism. ## Frequently Asked Questions ### Does UniProt have expression and localization data for a species that is not a model organism? Often yes, but usually through inference rather than direct experiment. Any species with a genome assembly can have a UniProt reference proteome, and unreviewed entries carry automated annotations including subcellular location. Direct, experimentally supported expression and localization annotation is concentrated in species with substantial curated literature, so verify evidence tags on every claim. ### What is the difference between reviewed and unreviewed UniProt entries? Reviewed entries carry manual annotations produced by expert curators from published experiments. Unreviewed TrEMBL entries carry annotations generated by automated, rule-based systems. Functionally, this means a reviewed localization claim traces to an observation, while an unreviewed one often traces to a computational inference from a related protein. ### How do I download all subcellular localization annotations for my species in bulk? Use the UniProt REST API with a query filtering on your organism's taxon identifier and request the subcellular location and tissue specificity fields in TSV format. The command pattern shown earlier in this article retrieves an entire species' annotations in one request, ready for joining to your gene identifiers through the ID mapping service. ### How reliable are annotations transferred from orthologs in non-model organisms? Reliability scales with orthology confidence and phylogenetic distance. Close, one-to-one orthologs with reviewed source entries are strong starting points. Distant or many-to-many relationships produce weaker inferences, and multi-species reference strategies give more accurate annotation than reliance on any single model organism. ### Should I use UniProt, Expression Atlas, or MOPED for expression data? Use UniProt for the annotation frame and evidence tags, Expression Atlas for measured gene and protein expression across tissues and conditions where your species or a close relative is covered, and MOPED for proteomic detection records in humans and model organisms. They answer different questions, and a sound workflow chains them rather than choosing one. ### How should I report UniProt-derived localization claims in a paper or grant application? State the source entry accession, the evidence type behind the claim, and whether the annotation is experimental or inferred by similarity or by automated rule. If the claim is central to your hypothesis, cite the primary experimental paper embedded in the UniProt entry rather than the entry alone, and describe your validation plan for your own organism. ## Sources 1. [UniProt: a worldwide hub of protein knowledge](https://pubmed.ncbi.nlm.nih.gov/30395287/) 2. [Multiple model species selection for transcriptomics analysis of non-model organisms.](https://pubmed.ncbi.nlm.nih.gov/30367568/) 3. [Finding biomarkers in non-model species: literature mining of transcription factors involved in bovine embryo development.](https://pubmed.ncbi.nlm.nih.gov/22931563/) 4. [Expression Atlas in 2026: enabling FAIR and open expression data through community collaboration and integration](https://pubmed.ncbi.nlm.nih.gov/41370097/) 5. [MOPED: Model Organism Protein Expression Database](https://pubmed.ncbi.nlm.nih.gov/22139914/) 6. [Data on SVCT2 transporter expression and localization in cancer cell lines and tissues](https://pubmed.ncbi.nlm.nih.gov/31249848/) 7. [Experimental supporting data on TKS5 and Cortactin expression and localization in human pancreatic cancer cells and tumors](https://pubmed.ncbi.nlm.nih.gov/30581916/) 8. [Gene Tagging Strategies To Assess Protein Expression, Localization, and Function in Drosophila](https://pubmed.ncbi.nlm.nih.gov/28978772/) 9. [Semi-quantitative Determination of Protein Expression using Immunohistochemistry Staining and Analysis: An Integrated Protocol](https://pubmed.ncbi.nlm.nih.gov/31867411/) 10. [Breast Cancer Candidate Gene Detection Through Integration of Subcellular Localization Data With Protein-Protein Interaction Networks](https://pubmed.ncbi.nlm.nih.gov/32340955/) 11. [LABRAT reveals association of alternative polyadenylation with transcript localization, RNA binding protein expression, transcription speed, and cancer survival](https://pubmed.ncbi.nlm.nih.gov/34174817/) 12. [Expression, Localization, and Activity of the Aryl Hydrocarbon Receptor in the Human Placenta](https://pubmed.ncbi.nlm.nih.gov/30486367/) 13. [Normalizing and denoising protein expression data from droplet-based single cell profiling](https://pubmed.ncbi.nlm.nih.gov/35440536/)