# Understanding Variant Consequences in Ensembl In Ensembl, a variant consequence is the predicted effect of a DNA change on a specific transcript or regulatory feature, recorded as a controlled term from the Sequence Ontology such as missense_variant or splice_donor_variant. Because one variant can overlap several transcripts, Ensembl reports one consequence per affected feature rather than a single label per variant. That transcript-aware design is what makes the annotation useful and, at first, confusing. A single G to A change can be a silent change in one transcript, a missense change in a second transcript, and a splice-disrupting change in a third. Anyone working with sequencing data, whether from a rare disease exome, a cancer panel, or a population survey, needs to read these annotations correctly before prioritizing variants. This article explains how Ensembl assigns variant consequences, how to generate them with the Ensembl Variant Effect Predictor (VEP), how to read them in the [Ensembl genome browser](/blog/guides/ensembl-genome-browser), and how to move from a consequence label to a defensible interpretation. The Ensembl VEP is a freely available, open-source tool that predicts variant molecular consequences using the Ensembl/GENCODE or RefSeq gene sets, and it is updated roughly quarterly to incorporate new gene, variant, and phenotype information [1]. ## What Does "Variant Consequences Ensembl" Mean? When you search a variant identifier such as an rsID in the Ensembl genome browser, the variant page contains a section titled something like "Your genome" and, above it, a table of variant consequences. That table lists, for every transcript overlapping the variant, the transcript identifier, the biotype, the consequence term or terms, the codon and amino acid change where relevant, and a colored impact flag. This is what people mean when they talk about variant consequences Ensembl output. It is annotation, not diagnosis. It is a systematic prediction of where a change sits within the transcript architecture of the reference genome and what that position implies at the level of DNA, RNA, and protein. Two concepts underpin the whole system. First, consequences are drawn from the Sequence Ontology (SO), a formal, machine-readable vocabulary of biological sequence features and their relationships. Ensembl uses SO terms so that a missense_variant means exactly the same thing in Ensembl, in VEP output, in ClinVar submissions, and in downstream scripts. Terms can be combined with an ampersand when a variant overlaps two features at once, for example splice_acceptor_variant&intron_variant. Second, consequences are assigned per transcript, not per gene. Ensembl and GENCODE (the joint Ensembl and HAVANA annotation project) define multiple transcripts for most human genes, and RefSeq defines yet another set. A variant's effect depends on which transcript you ask about, which is why VEP and the browser always report the transcript identifier alongside the consequence. ### The core vocabulary The table below covers the terms you will encounter most often when reading variant consequences Ensembl output, with the impact category VEP assigns to each. Impact categories (HIGH, MODERATE, LOW, MODIFIER) are explained in their own section below. | SO term | Plain-language meaning | Typical molecular effect | VEP impact | |--|--|--|--| | missense_variant | Changes one amino acid to another | [Protein sequence](/blog/guides/protein-sequence) change of uncertain degree | MODERATE | | synonymous_variant | Changes a codon without changing the amino acid | Usually no protein change | LOW | | stop_gained | Introduces a premature stop codon | Truncated or nonsense protein | HIGH | | frameshift_variant | Insertion or deletion not divisible by three in coding sequence | Shifts the reading frame downstream | HIGH | | inframe_insertion / inframe_deletion | Adds or removes amino acids without shifting the frame | Alters protein length | MODERATE | | start_lost | Alters the annotated start codon | May abolish translation initiation | HIGH | | stop_lost | Alters the normal stop codon | Extends the protein into unknown sequence | HIGH | | stop_retained_variant | Changes the stop codon to another stop codon | No protein change | LOW | | protein_altering_variant | A coding change with complex or incompletely defined protein effect | Variable | MODERATE | | splice_donor_variant | Alters the invariant dinucleotide at the 5′ end of an intron | Likely disrupts splicing | HIGH | | splice_acceptor_variant | Alters the invariant dinucleotide at the 3′ end of an intron | Likely disrupts splicing | HIGH | | splice_region_variant | Lies within 1 to 3 bases of the exon or 3 to 8 bases of the intron near a splice site | Possible subtle splicing effect | LOW | | intron_variant | Falls inside an intron, away from splice motifs | Usually neutral | MODIFIER | | 5_prime_UTR_variant / 3_prime_UTR_variant | Falls in untranslated regions | May affect translation or [mRNA stability](/knowledge/molecular-biology/mrna-stability) | MODIFIER | | upstream_gene_variant / downstream_gene_variant | Within 5 kb of a gene, outside the transcript | Possible regulatory effect | MODIFIER | | intergenic_variant | Between genes | No direct effect on annotated transcripts | MODIFIER | | regulatory_region_variant | Falls in a regulatory feature | May affect regulation | MODIFIER | A few of these terms reward closer attention. protein_altering_variant appears when the protein consequence is real but does not fit a neat category, such as an in-frame deletion that both removes residues and disrupts a known domain. splice_region_variant is deliberately mild. It flags proximity to a splice site without claiming the conserved GT or AG dinucleotide itself is broken, and it should never be read as equivalent to splice_donor_variant or splice_acceptor_variant. The 5 kb window for upstream_gene_variant and downstream_gene_variant reflects VEP's default distance parameter, which can be adjusted at the command line. ## How Ensembl Assigns Variant Consequences, Step by Step ![Five step flow from placing the variant on an assembly to attaching supporting annotations such as HGVS and ClinVar.](/article-images/illustrations/variant-consequences-ensembl-vep-assignment-steps.webp) Understanding the assignment procedure turns the output from a wall of terms into something you can audit. Whether the annotation comes from the browser or from VEP, the logic is the same. **Step 1. Place the variant on the reference assembly.** The annotator needs an assembly and a coordinate system. Human annotation currently centers on GRCh38, with GRCh37 still supported for older projects. A variant is defined by chromosome, position, reference allele, and alternate allele. If your coordinates were generated against a different assembly than the one you annotate against, every downstream consequence is wrong, so assembly checking is the first habit to build. **Step 2. Find every overlapping feature.** The annotator queries the annotation database for all transcripts, exons, and regulatory features that overlap the variant position. This is where transcript multiplicity enters. A variant in a gene with five transcripts can produce five rows of output, plus a row for a nearby regulatory feature. **Step 3. Compute the change at each level.** For a transcript overlapping the variant, the annotator works out the effect on genomic DNA, then on the mature mRNA (the cDNA), then on the coding sequence (CDS), then on the protein. A change in the last base of an exon, for instance, is exonic at the DNA level but sits immediately adjacent to the splice donor site, and the annotator records both facts as combined terms. **Step 4. Assign the SO term or terms.** Based on the computed effect, the annotator selects the matching Sequence Ontology term. Combined terms are common. splice_donor_variant&intron_variant means the variant breaks the donor dinucleotide and lies in an intron. A frameshift near the 3′ end of a gene may also be annotated with a downstream term if the new reading frame runs into the next feature. **Step 5. Rank the impact.** Each SO term maps to one of four impact categories. HIGH means the term is usually expected to abolish [protein function](/blog/guides/protein-function) or truncate the transcript, such as stop_gained or frameshift_variant. MODERATE means a non-conservative but non-truncating change, the classic case being missense_variant. LOW means a change unlikely to alter the protein, such as synonymous_variant. MODIFIER means the variant does not fall in coding sequence or an essential splice signal, such as intron_variant or intergenic_variant. Impact is a severity ranking of the consequence class, not a probability of disease. **Step 6. Attach the supporting annotations.** Alongside the consequence term, VEP and the browser report protein position, the reference and alternate amino acids, codon change, HGVS genomic, coding, and protein nomenclature, existing database identifiers such as dbSNP rsIDs, known phenotype associations from ClinVar, allele frequencies from gnomAD and other studies, and deleteriousness predictions from tools such as SIFT and CADD [1]. This supporting layer is what converts a coordinate into something you can actually interpret. ### Which transcript set is being used VEP can annotate against the Ensembl/GENCODE gene set or against RefSeq transcripts, and this choice changes output [1]. The two projects differ in transcript models, particularly at gene boundaries and for genes with complex [alternative splicing](/blog/guides/alternative-splicing). Ensembl marks a canonical transcript for each gene, intended as a representative transcript, and MANE Select transcripts, agreed jointly by Ensembl and NCBI RefSeq, identify one matched transcript per gene across both resources. For clinical reporting, the choice of transcript must be explicit and consistent, because the same variant can carry missense_variant on one transcript and synonymous_variant on another. ## Annotating Your Own Variants With Ensembl VEP The Ensembl Variant Effect Predictor is the tool that produces these annotations at scale. It is open source, runs locally or on the Ensembl servers, and exposes three access routes designed for different levels of experience and different data sizes: a configurable command-line tool, a REST application programming interface, and a user-friendly web interface [1]. A machine learning study of ClinVar variants illustrates how standard the tool has become. That framework trained classifiers on 104,646 high-confidence ClinVar germline variants annotated with Ensembl VEP version 114 on GRCh38 together with CADD scores, achieving an area under the ROC curve of 0.9995 with a [random forest](/blog/data-analysis/what-is-random-forest) model before reclassifying tens of thousands of variants of uncertain significance [2]. The numbers belong to that specific model, but the design choice is the message for a workflow builder. VEP output is treated as a stable, feature-rich substrate for downstream analysis. ### Choosing an access method ![Decision path asking about variant count and pipeline needs, ending at the web interface, command line or REST API.](/article-images/illustrations/variant-consequences-ensembl-access-route-choice.webp) | Route | Best for | Practical notes | |--|--|--| | Web interface | Small variant sets, quick checks, learning the output | Handles upload of modest files, offers configurable options through menus, no installation [1] | | Command line (local install) | Large VCFs, pipelines, plugins, custom annotations | Requires Perl and a downloaded cache, supports plugins and custom GFF transcript sets | | REST API | Programmatic lookups of small numbers of variants | Useful for web applications and scripted spot checks | The web route is genuinely capable. The published VEP tutorial from the Ensembl team walks through sophisticated analysis using the web tool alone, with the explicit aim of making the tool accessible to users without command-line experience [1]. If you are annotating a 30-variant Sanger validation list, start there. If you are processing thousands of exomes, install the command-line version with a local cache, because repeated web queries at that scale are neither practical nor polite to a shared service. ### Inputs and a minimal run VEP accepts VCF, a simple tab-delimited default format, region strings, and HGVS notation. The VCF route is the most common in practice because variant callers emit VCF directly. A minimal input line looks like this, tab-separated, with the header included: ``` ##my-vcf-header-lines #CHROM POS ID REF ALT QUAL FILTER INFO 7 140453136 rs11568868 A G . PASS . ``` With a local cache installed, a typical command-line run names the input file, the output file, and the options that shape the annotation. Rather than reproducing flag syntax here, the table below lists the options that matter most in everyday work, using the plain option names as they appear in the VEP documentation, so you can look up the exact syntax for your installed version. VEP is updated roughly quarterly, and exact flag behavior should be checked against the documentation for the release you have [1]. | Option (by name) | What it does | |--|--| | cache | Use the downloaded offline cache instead of a live database | | offline | Run with no network connection, required for fully reproducible pipelines | | assembly | Select the reference assembly, typically GRCh38 | | hgvs | Add HGVS genomic, coding, and protein nomenclature | | symbol | Add gene symbols alongside Ensembl identifiers | | canonical | Mark which transcript is the Ensembl canonical for each gene | | pick | Report a single best consequence per variant instead of all of them | | refseq | Annotate against RefSeq transcripts instead of Ensembl/GENCODE | | plugin | Attach external prediction and annotation tools such as CADD | | check_ref | Verify that the reference allele matches the assembly | Two habits separate reliable runs from unreliable ones. Always record the VEP version, the cache version, and the assembly with your results, because consequence calls can change as transcript models are updated between releases. And when you annotate RefSeq-based clinical data, use the RefSeq option deliberately rather than by accident, since mixing transcript sets across a project produces internal inconsistencies that are painful to debug later. ### Output fields worth reading A VEP tab-separated output row carries the consequence term in the Consequence column, the impact category in the IMPACT column, the gene symbol and identifier, the transcript identifier and biotype, the exon or intron number, HGVS strings, the amino acid change, and, when enabled, the prediction scores and frequency fields. The ClinVar phenotype column and the gnomAD frequency columns are the two additions most frequently requested in triage settings, and both are available through the appropriate options or plugins because VEP incorporates phenotype and frequency data as part of its design [1]. For pipeline work, structure matters as much as content. The Germline VCF Annotator project, built on VEP for a whole-genome sequencing [cohort study](/blog/guides/cohort-studies-a-practical-guide-to-design-conduct-and-interpretation), generated VEP tabular annotations with explicit allele fields so that multi-allelic sites could not be mis-parsed, and then appended read-evidence metrics with a rules-based quality class. That workflow reported near-perfect within-patient concordance across technical repeats at predefined [DNA damage response](/knowledge/molecular-biology/dna-damage-response) loci after filtering for nonsilent SNVs with read depth of at least 15, with discordance concentrated among low-quality loci [3]. The lesson for anyone building a variant review table is concrete. Keep one row per allele, keep fields explicit, and never paste raw VCF into spreadsheet software, because automatic type conversion can distort the data [3]. ## Reading Variant Consequences in the Ensembl Genome Browser If you are learning how to use Ensembl for the first time, the genome browser is the fastest way to build intuition, because it shows consequences in their genomic context. A practical sequence for a single variant works like this. 1. Go to the Ensembl website and enter the variant identifier, gene symbol plus position, or a raw coordinate such as chromosome, position, and alleles into the search box. 2. Open the variant page. The consequence table at the top lists each overlapping transcript with its identifier, biotype, consequence terms, codon and amino acid change, and a colored impact flag. Red marks high impact, which helps you scan a long table quickly. 3. Expand a transcript row to see details such as HGVS nomenclature, protein domain overlaps, existing database identifiers, and frequency data from population studies. 4. Use the region in browser view to see the variant in sequence context, with tracks for transcripts, regulatory features, and conservation. This view immediately shows why a variant carries, for example, both an exonic and a splice_region term. 5. Check the phenotype and frequency data on the page. Known ClinVar assertions and population frequencies turn a bare consequence into a starting point for interpretation. The browser and VEP are two faces of the same annotation logic, so a consequence you see in the browser will match a VEP run with equivalent options. The browser excels at explaining a variant you already care about. VEP excels at producing the annotation for a thousand variants you have not met yet. ## From Consequence to Interpretation: Prioritization Workflows A consequence term is a starting point. The interpretive layer is built by filtering and scoring, and several published workflows show the standard patterns. **Diagnostic filtering with gene-to-phenotype matching.** The G2P system (gene2phenotype) associates allelic requirement and mutational consequence at a locus with phenotypic terms, and the VEP-G2P plugin filters whole exome data against those curated panels. On the Developmental Disorders panel, VEP-G2P showed a sensitivity of 97.3 percent and a precision of 33 percent for de novo pathogenic genotypes, and a sensitivity of 81.6 percent with precision of 22.7 percent for inherited pathogenic genotypes [4]. The precision numbers are the honest part of the story. Consequence-aware filtering drastically shrinks a variant list, and the remaining candidates still require human judgment. **Scoring within VEP.** VEP can attach deleteriousness predictions from tools such as SIFT and CADD directly to its output [1], which lets you combine a structural consequence, such as frameshift_variant in a tumor suppressor, with a quantitative intolerance score in a single filtering pass. Published gene-level studies follow this exact pattern. One analysis of non-synonymous SNPs in the alternative lengthening of telomeres genes SMARCAL1, DAXX, and ATRX retrieved polymorphisms from the Ensembl database and then layered independent predictors including PROVEAN, PolyPhen2, Predict-SNP, PANTHER, and PMut to rank the most deleterious candidates before stability, conservation, and post-translational modification analyses [5]. The general lesson is that consequence classification and pathogenicity prediction are separate layers that reinforce each other only when combined deliberately. **Machine learning on consequence-aware features.** Beyond hand-set filters, consequence annotations feed classifiers. The hereditary cancer framework mentioned above used VEP-annotated features plus CADD to classify ClinVar variants, and it derived probability thresholds of at least 0.80 for the pathogenic class and at most 0.20 for the benign class from precision-recall analysis, with empirically validated precision above 99.6 percent for both classes on held-out data [2]. Reproducing those specific numbers requires reproducing that specific training set, but the architecture, VEP annotation in, prioritization out, is broadly reusable. **Custom transcriptomes.** Reference transcript sets, including Ensembl/GENCODE and RefSeq, miss transcripts that are expressed only under particular conditions, and variants affecting those transcripts escape correct annotation. SUsPECT addresses this by running VEP against custom transcript sets built from long-read [RNA sequencing](/blog/guides/rna-sequencing), predicting consequences and missense deleteriousness for novel open reading frames. The authors demonstrated that some ClinVar pathogenic variants show more severe predicted consequences under a stimulated immune-cell transcriptome than under the reference transcriptome [6]. If your tissue of interest has condition-specific isoforms, annotating against the reference alone can systematically under-call severity. **Protein-level context.** Consequence terms say nothing about which protein domains or modification sites are touched, and that context often decides whether a MODERATE missense matters. EXPANSION, a web server for cancer genomics, maps InterPro domains, post-translational modifications, and isoform-specific protein interactions from IntAct onto Ensembl transcripts to expose functional differences between splice isoforms [7]. A complementary obesity-gene study took a related path, mapping mutations through SnpEff against GRCh38.86, correlating Ensembl transcript identifiers with UniProtKB accessions, and pulling post-translational modification annotations from resources including PhosphoSitePlus and UniProt. Across 66 obesity-related genes and 3,581 mutations, functional annotation showed an excess of perturbing missense and frameshift variants in the BBSome, LEPR, GNAS, and MC4R genes, with ALMS1, POMC, and LEPR emerging as topological hubs in the integrated network [8]. **A single-variant worked example.** For a concrete end-to-end case, consider a gestational diabetes study that screened the choline acetyltransferase gene (CHAT), which encodes the enzyme responsible for acetylcholine synthesis and contributes to insulin secretion regulation. Sequence alignments and variant annotation were performed with UGENE software and Ensembl VEP. The team identified a novel heterozygous missense variant in exon 8, annotated as CHAT (NM_020549.5):c.1213C>G (p.Leu405Val). Interpretation rested on the consequence plus surrounding evidence, including a predicted haploinsufficiency mechanism, the location of the variant 18 bases in-frame from a known stop-gained variant (p.Gly411Ter), and regulatory data from RegulomeDB [9]. Notice the workflow shape. The consequence term opened the case. Protein and regulatory context carried the interpretation. ## Common Mistakes and Limitations Every one of these mistakes shows up in real projects, usually at review time. **Reading only the first output line.** VEP prints one line per transcript-feature overlap. Scripts or reviewers who grab the first row may read a modifier consequence from a short non-coding transcript while the clinically relevant high-impact row sits six lines down. Use the pick option only when single-transcript reporting is a deliberate, documented choice, and always record which transcript a reported consequence belongs to. **Treating IMPACT as pathogenicity.** IMPACT ranks the severity of the consequence class. A HIGH impact synonymous_variant, such as one that creates a new splice donor, is rare, but the general point stands. HIGH impact means the transcript-level change is predicted to be severe, not that the variant is disease causing. Population frequency from gnomAD and clinical assertions from ClinVar, both available within VEP [1], are the corrective evidence, and in silico predictions are known to perform inconsistently across contexts [2]. **Ignoring the transcript set.** Ensembl/GENCODE and RefSeq disagree on transcript models for many genes [1]. An interpretation memo that does not state which transcript set and which transcript identifier it used is incomplete, because the consequence itself can differ between sets. MANE Select transcripts exist precisely to give cross-database agreement for clinical work. **Mixing assemblies.** GRCh37 and GRCh38 coordinates differ, sometimes by large offsets on certain chromosomes. Annotating GRCh37 coordinates against a GRCh38 cache produces plausible-looking nonsense. Confirm assembly on both the input and the cache. **Overreading splice_region_variant.** splice_region_variant sits within 1 to 3 bases of the exon or 3 to 8 bases into the intron at a splice junction. It is a LOW impact flag for possible subtle effects. Confusing it with splice_donor_variant or splice_acceptor_variant, which mark the conserved dinucleotides themselves, inflates variant counts and wastes review time. **Assuming every intron_variant is silent.** Deep intronic variants can activate cryptic exons, and the reference annotation cannot see a transcript that is not in it. This is exactly the gap custom transcriptomes address, and the SUsPECT result shows reference-only annotation can miss severe effects entirely [6]. **Forgetting that annotation versions move.** VEP is updated roughly quarterly with new transcript models and phenotype data [1]. A consequence call from an old cache may not match the current browser. Record versions, and reannotate critical variants when you upgrade. **Building review tables by hand in spreadsheets.** Raw VCF and annotated outputs are machine formats. Spreadsheet import can silently corrupt variant fields through automatic type conversion, which is why structured, table-based workflows with explicit allele fields were built on top of VEP output in the first place [3]. **Expecting prediction to substitute for evidence.** Consequence prediction is one thread of variant interpretation. Clinical classification also needs segregation data, functional evidence, frequency, and phenotype match. Automated filtering tools exist to organize that evidence, and their published precision figures, such as the 33 percent precision for de novo genotypes in the VEP-G2P evaluation [4], make clear that human review of the filtered list remains mandatory. Individual diagnostic decisions also require input from qualified clinical genetics professionals. ## Quick Review - A variant consequence in Ensembl is the predicted effect of a change on a specific transcript or regulatory feature, expressed in Sequence Ontology terms. - Consequences are assigned per transcript, so one variant commonly carries several terms across its overlapping transcripts. - Impact categories (HIGH, MODERATE, LOW, MODIFIER) rank the consequence class, not the probability of disease. - VEP produces these annotations through a web interface, a command-line tool, and a REST API, against Ensembl/GENCODE or RefSeq transcripts, and is updated roughly quarterly [1]. - Always record the VEP version, cache, assembly, and transcript identifier attached to every reported consequence. - Filtering workflows combine consequence terms with frequency, ClinVar status, and scores such as CADD, and published evaluations show filtering is highly sensitive but leaves substantial human review [4]. - Reference transcript sets miss condition-specific isoforms, which custom transcriptome annotation can recover [6]. ## Frequently Asked Questions ### What does "variant consequences Ensembl" mean? It refers to the annotation Ensembl reports for how a sequence change affects each overlapping transcript or regulatory feature, using standard Sequence Ontology terms such as missense_variant or splice_donor_variant. You see it on Ensembl variant pages and in VEP output. ### Why does one variant show several different consequence terms? Because Ensembl annotates per transcript, and most genes have several transcripts with different exon structures. The same base can be coding in one transcript and intronic in another, so both consequences are true and both are listed. ### Is a HIGH impact consequence the same as a pathogenic variant? No. HIGH impact means the predicted transcript-level effect is severe, such as a frameshift or an altered splice donor. Pathogenicity also depends on gene relevance, frequency, clinical assertions, segregation, and functional evidence, which consequence terms alone do not supply. ### How do I annotate my VCF with Ensembl VEP? Install VEP with the appropriate cache for your assembly, run it on the VCF with the cache and offline options enabled, and enable options such as hgvs, symbol, and canonical depending on what your downstream table needs. For small variant sets, the web interface handles the same annotation without installation [1]. ### Should I use Ensembl or RefSeq transcripts for annotation? VEP supports both, and the choice changes transcript models and therefore consequences [1]. Match the transcript set your collaborators or clinical standards use, state it explicitly in methods, and stay consistent across the project. MANE Select transcripts give a single agreed transcript per gene across both systems. ### Can VEP annotate against transcripts that are not in the standard reference sets? Yes. VEP can be run against custom transcript sets supplied as GFF or GTF annotations, which is the approach used by pipelines such as SUsPECT for long-read transcriptomes, and it can reveal severe consequences invisible to reference-only annotation [6]. ## Sources 1. [Annotating and prioritizing genomic variants using the Ensembl Variant Effect Predictor-A tutorial.](https://pubmed.ncbi.nlm.nih.gov/34816521/) 2. [A machine learning framework for predictive interpretation of variants of uncertain significance in hereditary cancer.](https://pubmed.ncbi.nlm.nih.gov/42614771/) 3. [Germline VCF Annotator: a lightweight pipeline for processing germline VCFs with robust variant extraction and read evidence quality control.](https://pubmed.ncbi.nlm.nih.gov/41993465/) 4. [Flexible and scalable diagnostic filtering of genomic variants using G2P with Ensembl VEP.](https://pubmed.ncbi.nlm.nih.gov/31147538/) 5. [Investigating the structural and functional consequences of germline single nucleotide polymorphisms located in the genes of the alternative lengthening of telomere (ALT) pathway.](https://pubmed.ncbi.nlm.nih.gov/39021990/) 6. [SUsPECT: a pipeline for variant effect prediction based on custom long-read transcriptomes for improved clinical variant annotation.](https://pubmed.ncbi.nlm.nih.gov/37280537/) 7. [EXPANSION: a webserver to explore the functional consequences of protein-coding alternative splice variants in cancer genomics.](https://pubmed.ncbi.nlm.nih.gov/37810457/) 8. [Systems biology-driven analysis of obesity-associated genes with transcript mapping and post-translational modification annotations.](https://pubmed.ncbi.nlm.nih.gov/42172902/) 9. [The identification of novel missense variant in ChAT gene in a patient with gestational diabetes denotes plausible genetic association.](https://pubmed.ncbi.nlm.nih.gov/40688037/)