From Legacy Databases to Modern Standards: A Bioinformatics Pipeline for Migrating Immunodeficiency Variant Data to LOVD 3.0 with GRCh38 Standardisation

Abstract

Primary immunodeficiency diseases (PIDs) are a heterogeneous group of inherited immune system disorders caused by pathogenic variants in over 400 genes, collectively affecting an estimated one in 10,000 individuals worldwide. The IDbases collection, maintained on the legacy MUTbase platform at Lund University, represents one of the most comprehensively curated repositories of such variants, cataloguing disease-causing variants across 136 immunodeficiency-related genes with 6,259 patient records accumulated over more than three decades of expert curation. Despite their scientific value, these databases remain in a format incompatible with current genomic standards: variant coordinates are anchored to obsolete partial ENA clone reference sequences, nomenclature does not conform to Human Genome Variation Society (HGVS) recommendations, and the data cannot be integrated with modern interoperable platforms.
This thesis presents a systematic seven-step bioinformatics pipeline for migrating the IDbases data into the Leiden Open Variation Database 3.0 (LOVD 3.0), the leading open-source platform for locus-specific variant databases. The pipeline extracted variants from all 136 gene databases, resolved reference sequence identities through an empirical offset algorithm, lifted over chromosomal coordinates to GRCh38, and normalised all variant descriptions to HGVS nomenclature through three parallel Mutalyzer 3 annotation tracks. Track A (NG_IDRefseq genomic route) achieved a 99.3% normalisation success rate (3,582/3,606 variants). Track B (NM_MANE) achieved 94.1% (4,496/4,775). Track C (NM_IDRefseq) served as a tertiary traceability fallback at 59.5% (2,862/4,811).
A central finding of this work is that IDbases IDRefSeq sequences are partial ENA clone fragments contained within full LRG NG_ reference sequences - a previously undocumented structural relationship that causes systematic out-of-bounds coordinate failures for variants added after the original database construction date. The empirical offset algorithm developed to resolve this achieved a match rate of ≥90% for 92 of 94 testable genes (genes where an IDRefSeq FASTA was available) in the genomic track. Following priority-based merging of the three tracks, 7,776 patient–variant records were resolved and, after deduplication, correspond to 2,240 distinct variants across 115 genes, linked to 2,729 patients (5,773 patient–variant pairs). A further 698 distinct variants remained unresolved, of which roughly three-quarters are automatically rescuable, giving an automated resolution rate of 76.2% with an estimated addressable ceiling near 94%. Systematic data quality issues uncovered during the migration - including strand orientation inconsistencies in 43 reverse-strand genes, missing insertion sequences, legacy IVS intronic notation, and transcript version obsolescence - are catalogued as findings to inform future variant database design.
The pipeline is implemented as a series of documented, modular Python scripts and is designed for reuse in similar legacy database migration projects. The work makes a large, expertly curated immunodeficiency variant dataset available in a modern, interoperable format aligned with GRCh38 and MANE (Matched Annotation from the NCBI and EMBL-EBI) Select transcript standards.

Popular Abstract

Inherited immune diseases are rare, but for the people affected they can be severe: infections that are normally manageable may become repeated, dangerous, or hard to explain. In many cases, the clue is a small change in DNA. My thesis was about rescuing an old but valuable database of these changes, so that it can still be used with the tools researchers and clinicians rely on today.
For many years, researchers working with the IDbases collection recorded variants linked to primary immunodeficiency diseases. The collection covers 136 genes and thousands of patient records, which represents a large amount of careful expert work. The problem was not the biology; it was the format. The database had been built over a long period, using reference sequences and naming conventions that made sense at the time but no longer match current standards. Without a dedicated curator and regular updates, the data gradually became harder for modern software to read.
A variant database works a little like a map. To see where a DNA change is, you need to know which reference sequence the position is measured from. During the project, I found that many of the old IDbases references were not full gene references. They were shorter clone fragments sitting inside longer modern references. This matters because a position that is correct on a small fragment can become misleading when compared directly with a full-length reference. In some cases, variants recorded later in the history of the database were effectively being described outside the piece of sequence that had originally been used.
To solve this, I built a pipeline that finds where each old fragment belongs within the longer reference sequence. Once that right starting point is determined, the old coordinates can be shifted to that correct positions and then converted to GRCh38, the current human genome reference. Then Mutalyzer was used to rewrite and check the variants in HGVS nomenclature, the standard way of describing genetic variants.
The final output is a cleaned and standardised dataset ready for import into LOVD 3.0, an open variant database platform. After merging the different annotation routes and removing duplicates, the project resolved about 2,240 distinct variants across 115 genes, linked to roughly 2,700 patients.
The practical value is simple: information that had become difficult to use is now much closer to being searchable, shareable, and reusable. The project also leaves behind a reusable Python workflow for other legacy databases with similar reference-sequence problems. Not every record was rescued automatically, and some will still need manual checking, but the main structure is now in place.

• author: Nguyen, Ngoc Khoi
• supervisor: Mauno Vihinen
• organization: Biotechnology; Biotechnology (MSc)
• course: KMBM01 20261
• year: 2026
• type: H2 - Master's Degree (Two Years)
• subject: Technology and Engineering
• keywords: primary immunodeficiency, variant database migration, HGVS nomenclature, LOVD 3.0, GRCh38, Mutalyzer, IDbases, MUTbase, locus-specific database, bioinformatics pipeline, reference sequence standardisation, biotechnology
• language: English
• id: 9242187
• date added to LUP: 2026-08-11 08:56:25
• date last changed: 2026-08-11 08:56:25