BIOGENOVO

01 / 05

Omics

GenomicsTranscriptomicsMetagenomicsProteomicsMicrobiomicsMulti-Omics

01 OmicsGenomicsTranscriptomicsMetagenomicsProteomicsMicrobiomicsMulti-Omics

Featured Case Studies

CompletedArtificial Intelligence

PTC-Predict: Ribosomal Mutation Effect Prediction

Project Metadata

Project Type: Structural bioinformatics + ML web application
Domain: Ribosome biology / Mutation effect prediction
Objective: Predict functional effects of ribosomal mutations
Input: Ribosomal mutations + 3D structural features
Methods: Structural feature engineering; supervised ML; classification
Tools & Technologies: Python, BioPython, MDAnalysis/PyMOL, scikit-learn, XGBoost, SHAP, SQL, Web stack
Output / Deliverable: Mutation database + trained predictor + web application
Organism: Homo sapiens, Saccharomyces cerevisiae, Escherichia coli
Deployment / Platform: PTC-Predict web application

Background

Developed a structure-informed system to predict the functional effects of ribosomal mutations from their 3D structural context. The goal was to classify mutations as assembly disruption, functional impairment, antibiotic resistance, or neutral/tolerated.

The Challenge

Existing approaches often lack structural context and struggle with small, imbalanced datasets of experimentally characterized mutations. The challenge was to combine diverse structural and biochemical features into an interpretable predictive model.

Approach

Integrated PDB-derived structural feature extraction, machine-learning classification, mutation database design, and a web-based prediction interface. Structural features were analyzed with supervised ML and the resulting model was deployed as the PTC-Predict web application.

Impact

Validated the predictive pipeline using stratified cross-validation and held-out evaluation, with performance assessed using macro F1, precision, recall, accuracy, and ROC-AUC. The resulting platform enables researchers to explore known mutations and obtain structure-informed functional-effect predictions through a web interface.

CompletedOmics

Multi-Omics Pipeline for Aquatic Toxicity Response in Daphnia magna

Project Metadata

Project Type: Multi-omics
Domain: Toxicogenomics, Environmental Bioinformatics
Objective: Characterize molecular toxicity response & map pollution hotspots
Organism: Daphnia magna
Input: GSE7666, GSE7667, GSE74736; EU water quality dataset
Methods: Batch correction, DGE/DM analysis, multi-omics factor integration (MOFA2), ortholog mapping, functional enrichment, PPI network clustering, K-means geospatial clustering
Tools & Technologies: R, limma, MOFA2, OrthoDB, DAVID, KEGG Mapper, STRING, Python (K-means, heatmaps)
Deliverable: MOFA2 factor analysis report, network and enrichment report, geospatial pollution maps

Background

The project addresses how the aquatic sentinel species Daphnia magna responds molecularly to pollutant exposure, and where in Europe such exposure is most severe. It combines real-world geospatial pollution mapping with a multi-omics (transcriptomics + metabolomics) investigation of toxicity-induced molecular changes.

The Challenge

Key obstacles included integrating transcriptomic and metabolomic datasets with zero overlapping samples and differing sample sizes, working with a non-model organism lacking functional annotation, and correcting batch effects across microarray platforms (GSE7666/GSE7667) — plus extreme, non-normal variability in pollutant concentrations across European water bodies (ranging from 0 to >10¹⁴ mg/L in some lakes/rivers).

Approach

Built an end-to-end R-based pipeline: limma-based background correction and normalization of raw GPR microarray data, ComBat batch correction, limma differential expression/metabolite analysis, and MOFA2 for unsupervised latent-factor integration of the two omics layers. Top factor-loading features were mapped to Drosophila melanogaster orthologs (OrthoDB) for DAVID/KEGG enrichment and STRING PPI clustering. Separately, K-means clustering and correlation heatmaps were applied to geospatial pollution data across five European water body types.

Impact

Identified 456 significant metabolites (FDR<0.05) and 176 significant DEGs (adj. p<0.05) between exposed and control groups; MOFA2 Factor 2 was significantly associated with treatment (p<0.0001, effect size = 1.83) and orthogonal to other factors, confirming a robust, non-redundant toxicity signature. Downstream enrichment revealed a significant oxidoreductase-activity cluster (p=0.012, 16.65-fold enrichment) and 5 distinct functional protein clusters via PPI network analysis, while geospatial clustering pinpointed Greece and southeastern Europe as extreme contamination hotspots.

CompletedArtificial Intelligence

RNA-Binding Protein Prediction via ESM2 Embeddings and a Hybrid LSTM-GCN Ensemble

Project Metadata

Project Type: Supervised classification, deep learning
Domain: Computational biology
Objective: Predict RNA-binding proteins from sequence
Input: Protein sequences (ESM2 embeddings)
Methods: LSTM, GCN, hybrid fusion, ensemble, threshold tuning
Tools & Technologies: PyTorch, ESM2 (facebook/esm2_t30_150M_UR50D), scikit-learn
Output / Deliverable: Binary classifier (RNA-binding vs non-binding)

Background

Identifying RNA-binding proteins (RBPs) computationally from sequence/structure data is critical for understanding post-transcriptional regulation, yet experimental validation is slow and expensive. This creates a need for accurate in-silico classifiers that can screen proteomes for RNA-binding potential.

The Challenge

RBP datasets are heavily imbalanced (RNA-binding proteins are a small minority class), making single-architecture models prone to poor recall or false positives on the minority class. Sequence-only (LSTM) or structure-only (GCN) models each capture only part of the signal, and generalization to unseen species/organisms is often weak.

Approach

Built individual LSTM and GCN models using ESM2 (facebook/esm2_t30_150M_UR50D) embeddings as input features, then combined them into a hybrid and ensemble architecture to leverage both sequence and graph-structural information. Applied optimal threshold tuning (via precision-recall optimization) and validated the pipeline with 5-fold cross-validation, plus external evaluation on independent E. coli and Salmonella test sets.

Impact

The ensemble model achieved 98.09% accuracy, 0.735 F1-score on RNA-binding class, 0.726 MCC, and 0.983 AUC on the held-out test set. Cross-validation confirmed stability (Test Accuracy: 0.9798 ± 0.0010, AUC: 0.9244 ± 0.0110), and external validation on Salmonella reached 94.22% accuracy with 0.964 AUC, demonstrating strong cross-species generalization.

CompletedOmics

Multi-Omics Integration Reveals Dose-Dependent Chemical Drivers of Stress Response in Daphnia magna

Project Metadata

Project Type: Multi-omics data integration
Domain: Environmental toxicology
Objective: Identify chemical drivers of stress in Daphnia
Organism: Daphnia magna
Dataset / Scale: 144 samples, 13 sites, 3 doses
Input: RNA-seq, metabolomics, water chemicals
Methods: PCA, WGCNA, RGCCA, iRF, LOOCV
Tools: R, WGCNA, RGCCA, iRF, OrthoDB, KEGG/HMDB
Output: Ranked chemical drivers, stress modules

Background

The project is a multi-omics study of Daphnia magna exposed to water samples across 13 locations (control + D01–D12) and 3 concentration levels (control, 1x, 10x), combining RNA-seq (18,903 genes → 13 WGCNA module eigengenes), metabolomics (3,616 features), and water chemical profiling (91 chemicals measured; the plan references 38 as predictors).

The Challenge

Two research goals drove the work: (1) understanding how gene expression and metabolite changes are inter-connected, and (2) pinpointing which specific chemicals among many are the “smoking gun” driving biological stress — all while ruling out that results are simply confounded by sampling location rather than true chemical exposure.

Approach

A five-phase pipeline was used: sample alignment and PCA for QC, WGCNA to reduce genes/metabolites into co-expression modules, RGCCA to integrate RNA modules with metabolomics across 3 components, iRF (via LOOCV) to rank the 91 water chemicals as predictors of each stress module and detect chemical interactions, and final biological/chemical annotation using ortholog and KEGG/HMDB mapping.

Impact

Concentration (not location) was found to be the dominant driver, with RNA and metabolomics strongly co-regulated on Component 1 (r ≈ 0.687, p < 0.001, shared AVE = 0.203) and no geographic confounding across 12 sites. Acesulfame and Carbamazepine emerged as universal wastewater-contamination markers present across all four stress modules, with Diphenylphosphate and Oxazepam flagged as high-priority toxicological drivers — collectively implicating wastewater effluent as the primary source of chemical stress on Daphnia.

CompletedComputational Development

LabAssist: LLM Chatbot for Rapid Laboratory Procedure Retrieval

Project Metadata

Project Type: LLM / RAG Chatbot
Domain: Laboratory Informatics / AI
Objective: Rapid laboratory procedure retrieval
Input: CRISPR protocol PDF/DOCX files
Methods: Document ingestion; text chunking; vector retrieval; RAG
Tools & Technologies: Python; LangChain; Ollama; Llama 3 ChatQA 8B; ChromaDB; PyMuPDF; docx2txt
Retrieval: Top-3 semantic document retrieval
Embeddings: Ollama mxbai-embed-large
Output / Deliverable: Conversational protocol answers with source documents
Deployment / Platform: Local Ollama runtime
Team Role: Backend Developer
Event: St. Jude KIDS25 BioHackathon

Background

Developed collaboratively at the St. Jude KIDS25 BioHackathon to provide laboratory personnel with a conversational interface for quickly accessing CRISPR experimental protocols and procedures.

The Challenge

Laboratory protocols are distributed across PDF and Word documents, making specific procedural information cumbersome to locate and reference during experimental work.

Approach

Built a local LLM-powered RAG chatbot using LangChain, ChromaDB, Ollama embeddings, and Llama 3 ChatQA 8B; PDF/DOCX protocols were chunked, indexed, retrieved, and supplied as context to the LLM.

Impact

Created a privacy-friendly, locally running conversational knowledge interface that answers protocol questions from retrieved source documents and identifies the documents supporting each response.

In DevelopmentArtificial Intelligence

Genomic Variant Pathogenicity Prediction Using a Hybrid GCN-BiLSTM

Project Metadata

Project Type: Deep learning prototype
Domain: Clinical genomics
Objective: Classify variants as harmful or non-harmful
Input: Labelled variant CSV, hg38 FASTA
Methods: GCN + attention BiLSTM fusion, k-mer encoding, focal loss
Tools & Technologies: Python, PyTorch, PyTorch Geometric, NetworkX, pyfaidx, scikit-learn
Output / Deliverable: Trained model, training curves, test metrics
Organism: Human
Reference / Genome: hg38

Background

A model was needed that separates harmful from non-harmful human genetic variants, with gene-locus annotation of disease mutations.

The Challenge

Harmful variants make up only about 22% of the labelled data. Each prediction must combine the local sequence change around a variant with gene-level context.

Approach

Variants were filtered to single-character alleles. For each, a 21-bp hg38 window was extracted in reference and mutated forms plus reverse complements, then 3-mer one-hot encoded. A gene graph (13,398 nodes, 40,794 edges) feeds a 2-layer GCN, and a 3-layer attention BiLSTM encodes the four sequences. The two are fused in an MLP head and trained with focal loss on a stratified 70/15/15 split, with early stopping.

Impact

The model trained on GPU with best validation loss 0.0380. On the held-out test set it reached accuracy 0.858, precision 0.945, recall 0.868 and F1 0.905.

CompletedPrecision Medicine

Comprehensive Pharmacogenomic Profiling with PyPGx, Aldy, CNVnator and VEP

Project Metadata

Project Type: Pharmacogenomic Analysis
Domain: Precision Medicine / Drug Response
Inputs: VCF, BAM/BAI, GRCh38 reference FASTA
Methods: Star-allele calling · Haplotype analysis · CNV analysis · Variant annotation
Tools: PyPGx, Aldy, CNVnator, VEP
Reference: GRCh38
Deliverables: PGx profiles, Star alleles, CNV/SV calls, Functional annotations
Documentation: QC · Logs · Commands · Software versions · Checksums

Background

Develop a comprehensive PGx profile from sequencing data to identify pharmacogenetic variants, star alleles, copy-number changes, and annotations relevant to drug response and toxicity.

The Challenge

PGx interpretation requires coordinated analysis across variant calls, read-level evidence, haplotypes, CNVs, gene annotations, and reference resources while maintaining complete reproducibility.

Approach

PyPGx and Aldy are run across all genes supported by each tool, while CNVnator uses the curated PGx BED for region-specific CNV analysis. VEP provides downstream functional annotation of detected variants.

Impact

A documented PGx analysis package containing allele/haplotype calls, CNV results, functional annotations, summary tables, reports where supported, execution logs, exact parameters, software/reference versions, and checksums.

CompletedPrecision Medicine

Clinical Variant Curation and Prioritization for Pharmacogenomic Candidates

Project Metadata

Project Type: Clinical variant curation
Domain: Human genomics, pharmacogenomics
Objective: Prioritise disease-relevant variants
Input: Annotated variant table, CRAM/CRAI
Methods: IGV review, MANE re-annotation, ClinVar/gnomAD lookup
Tools & Technologies: IGV, Franklin, VEP, ClinVar, gnomAD, dbSNP
Output / Deliverable: Annotated Excel/CSV

Background

Rare and potentially functional variants were assessed to link them to immune dysregulation, neurological and sensory hypersensitivity, mitochondrial and redox pathways, JAK/STAT signalling, channelopathies and drug sensitivity.

The Challenge

Candidates had to be separated from sequencing artefacts through manual read-level IGV review, especially the LILRB2 frameshifts and the G6PD calls. Transcript choice also changed interpretation: the original MSTO1 call came from a non-MANE transcript and appeared as a start-loss.

Approach

Variants were annotated against MANE Select transcripts and enriched with ClinVar, gnomAD and the in-silico predictors listed above. They were then checked in IGV and given a consensus classification, ACMG-style codes and a level of interest.

Impact

The working file is a prioritised, annotated table. Several variants are flagged Investigate = Y, including LILRB2, NDUFS3, SIAE and PRKCG, and IGV review rejected some calls, such as the G6PD c.595A>G false positive.

In DevelopmentFunctional Genomics

Interactive RShiny Dashboard for Mouse Gene-Phenotype Visualization

Project Metadata

Project Type: Data handling and Visualization
Domain: Mouse functional genomics
Objective: Queryable DB and dashboard of IMPC phenotypes
Input: IMPC per-analysis .csv files (p-values); Phenodigm disease scores; SOP and parameter description files
Methods: Data cleaning; tidy formatting; relational normalisation; keyword-based parameter grouping; k-means with PCA (disease clustering in draft script)
Tools & Technologies: R, tidyverse, Shiny, plotly, MySQL, DBI/RMySQL
Output / Deliverable: R scripts; MySQL database and dump; RShiny dashboard; GitHub repository; individual report

Background

The IMPC knocks out individual mouse genes and compares phenotypes against wild-type controls. It reports linear-mixed-model p-values, FDR-adjusted, with a significance threshold of < 0.05. A collaborator requested these results as a queryable database and an interactive dashboard, linked to procedure, parameter and human disease information.

The Challenge

Raw per-analysis files had to be cleaned for accuracy, completeness and consistency, with missing data explicitly handled. They then had to be restructured into a normalised, scalable MySQL schema. Parameters also had to be grouped by test and naming similarity to reduce the parameter space, with weight, image and brain groups required plus at least three more.

Approach

Planned pipeline: retrieve data from the TRE, clean and collate it in R as tidy data, and build the MySQL database. The database has a draft ERD with Analysis, Phenotype, Procedure, parameter_group and Human_Disease tables. Parameters are grouped by keyword lists (heart, eye, electrolytes, FBC and lipid/glucose terms have been drafted). Finally, the data is served through an RShiny dashboard with p-value bar plots for selected genes or phenotypes and a clustering view.

Impact

The source material shows drafted deliverables: a database schema, parameter-group keyword lists and a working-draft dashboard script. The dashboard queries the MySQL database directly and flags p-value < 0.05 in the bar plots.

CompletedFunctional Genomics

Comparative Functional Analysis of Vector Integration Site Genes in iPSCs and Cardiomyocytes

Project Metadata

Project Type: Functional analysis
Domain: Gene therapy vector safety
Objective: Compare PHR vs PHV integration targets
Input: IS gene lists
Methods: Overlap analysis, Pathway enrichment, Cancer-gene classification, PPI network analysis
Tools & Technologies: Excel, GeneVenn, Reactome (v3.7), UniProt, STRING v12.0
Output / Deliverable: Overlap tables, 4 Venn diagrams, Top-10 pathway tables, Oncogene/TSG table, PPI networks with hub genes

Background

Integrating viral vectors can activate or disrupt nearby genes (insertional mutagenesis), which is a key safety concern in gene therapy. This project compares integration targets of a SIN-LTR vector (PHR) and a full-length-LTR control (PHV) in iPSCs and in cardiomyocytes after differentiation.

The Challenge

Each dataset is a gene list of thousands of targets, so shared and unique integration patterns, affected pathways and cancer-gene content had to be summarised across four conditions.

Approach

IS gene lists were organised in Excel to calculate total, unique and shared genes with overlap percentages, then visualised as pairwise Venn diagrams in GeneVenn. Reactome was used for pathway enrichment (top 10 per dataset), UniProt for oncogene and tumour-suppressor classification, and STRING for PPI networks of common IS genes in iPSCs and CMs, with hub genes highlighted.

Impact

1,441 genes are common to all four datasets, and PHV targets more genes than PHR in iPSCs (6,690 vs 4,008). RHO GTPase, MECP2/epigenetic and synaptic pathways recur across datasets, while oncogenes account for roughly 20–22% of IS genes and tumour suppressors roughly 6–7% in every dataset. Distinct hub-gene sets were identified for iPSCs (kinesin and cadherin genes) and CMs (dystrophin-associated complex genes, plus PBRM1).

CompletedPrecision Medicine

Targeted SCN9A Variant Calling, Annotation and Interpretation

Project Metadata

Project Type: Variant Calling
Domain: Clinical genomics / neuropathy genes
Objective: SCN9A variant calling, annotation, reporting
Input: Targeted-panel BAM + BAI
Methods: Region-restricted calling, database annotation, rule-based interpretation
Tools & Technologies: samtools, bcftools, tabix, ANNOVAR, Python (pandas), Ensembl REST API
Output / Deliverable: Annotated VCF, Excel variant summary
Expertise Tags: Variant calling, ANNOVAR annotation, ClinVar/gnomAD/dbSNP, coordinate liftover, variant reporting
Organism: Human
Dataset / Scale: Single sample, SCN9A region (~1 Mb), 27 variants

Background

Variants needed to be called and interpreted in the SCN9A region of a targeted-panel BAM. The deliverables were an annotated VCF and a short report covering c./p. nomenclature, allele frequency, clinical status and whether functional validation is needed.

The Challenge

The BAM was aligned to hg19, so calling and annotation had to use hg19 and the report had to state the reference version. The hg38 window therefore had to be converted to the matching hg19 coordinates.

Approach

hg38 coordinates were converted to hg19 coordinates with the Ensembl REST API. A scripted pipeline then called variants with bcftools, indexed the VCF, and annotated it with ANNOVAR using refGene, ClinVar, gnomAD and dbSNP (-vcfinput). A Python script parsed the annotated VCF into an Excel summary with position, gene, c./p. notation, allele frequency, ClinVar significance and a rule-based validation recommendation.

Impact

The pipeline produced an annotated VCF with 27 called variants and an Excel summary, both generated from a single reproducible Bash script and notebook.

CompletedComputational Development

Reproducible Germline Variant Calling and QC with DeepVariant

Project Metadata

Project Type: Variant Calling & QC
Domain: Genomic Analysis
Input: Aligned BAM/BAI
Output: VCF, gVCF, normalized/filtered VCFs, QC report & manifest
Tools: DeepVariant, samtools, bcftools, tabix
Reference: GRCh38

Background

Process an aligned human BAM against the exact GRCh38 reference used for alignment and generate a complete, archive-ready germline variant-calling package.

The Challenge

The delivery required strict reference consistency, pinned software versions, indexed variant files, reproducible QC statistics, and complete checksums to ensure the results could be independently verified and archived.

Approach

Applied a reproducible DeepVariant workflow covering variant calling, gVCF generation, normalization, filtering, indexing, and downstream QC, with reference, tool versions, commands, logs, and checksums documented.

Impact

Produced a traceable delivery package comprising raw and processed VCF/gVCF files, BAM/VCF QC statistics, README, manifest, and MD5/SHA256 checksums for reproducibility and archival.

Client Reviews

Trusted by researchers and developers worldwide

Real, unedited reviews from 58 clients on Fiverr — and 18 of them came back for more.

“Thanks for the delivery, we’re really happy with the quality of your work. The tutorial looks great, and we appreciate the extra effort you put into selecting the right dataset. We’d love to continue the collaboration. We’ll review everything in detail and reach out if we have any follow-up questions. Thanks again for the solid work.”

vincenthus974

Netherlands · Verified Fiverr client

5.0

“He delivered outstanding work with impressive attention to detail, surpassing my expectations. 🙌 Collaborating was a breeze thanks to their high level of cooperation and quick responsiveness. Highly recommend!”

joseph_xu

United States · Verified Fiverr client

“This isn't the first project we've done, and the feedback has been nothing but positive. Thank you for your work.”

migeram

Italy · Verified Fiverr client

Repeat client · 2 orders

“everything was perfect, he fixed all my problems! 10/10 would recommend!”

jakob_primer

Austria · Verified Fiverr client

“Seller is amazing. Completed complicated tasks perfectly and on time. Everything was done with precise accuracy and amazing attention to each detail. Seller has amazing skills within a field! Seller is also communicating and responding ASAP and is very helpful. I am beyond happy with his work 👌”

ajamex

United Kingdom · Verified Fiverr client

“If you need any problem solved in bioinformatics, this is the guy to go to. Extremely communicative, helpful, and kind. I had a tough task and he solved it with ease. Thank you so much!”

siguac

United States · Verified Fiverr client

“very smart et very professional!”

mika37

Belgium · Verified Fiverr client

Repeat client · 13 orders

“I am a novice in computer programing. He wrote a python script and delivered it to me and also showed me how to run the script for my bioinformatic-related work. I am really impressed with his effort to understand my needs with continuous back to back communication with him over few days”

arvindvirmani

United States · Verified Fiverr client

Repeat client · 5 orders

“Service was outstanding, good communication and the product was precisely what was asked. An excellent creative initiative was shown throughout the code. and the code itself was very well understandable with clear comments. Sidenote he shows a good understanding of basic bioinformatic principles. All in all, I certainly recommend his services.”

daansteur

Netherlands · Verified Fiverr client

“I had an urgent task that I needed help with. The seller was responsive, extremely helpful and delivered within a few hours with exactly what I asked for. Very efficient and perfect delivery. Highly recommend!!! Will definitely use services again”

jacobgreen51

United Kingdom · Verified Fiverr client

Repeat client · 3 orders

Read all 104 reviews on Fiverr

Reviews are shown exactly as written by clients.

Research Collaboration

Have a Biological Research Challenge?

From omics data analysis and AI-driven research to custom workflows and bioinformatics software, BioGenOvo works with researchers to develop practical computational solutions tailored to their research needs.

Beyond Omics, Towards Intelligence.