Bioinformatics Tutorials
2 postsPlain-language explainers on the file formats, workflows, and concepts behind sequencing analysis — written for biologists.
How to Download Raw Sequencing Data (FASTQ) from a GEO Dataset
GEO stores processed data — the raw FASTQ reads live in SRA/ENA. This guide explains the GSE → SRA → ENA → FASTQ chain, why the RAW.tar rarely holds reads, and the one-hop shortcut to raw sequencing data for any GEO series (no sra-tools needed).
Read more →From FASTQ to Results: The File Formats Behind Every Sequencing Analysis
Running the command was never the hard part — knowing what flows through the pipeline is. A biologist’s field guide to FASTQ, BAM/CRAM, VCF, GTF/GFF, BED, count matrices and single-cell formats: what each file holds, where it comes from, and where it goes next.
Read more →Tool Guides
5 postsHow to get the most out of the free BioTransfer tools.
How to Find CAR-T Targets and Trials for Any Gene
Gene-keyed drug databases structurally miss CAR-T targets: GD2 is a glycolipid with no gene of its own, and antigens like GPC2 or B7-H3 look undrugged because they were never small-molecule targets. Here is why, and a free way to search CAR-T and engineered cell-therapy trials by gene or antigen — CD19, BCMA, GD2, B7-H3, GPC2, mesothelin and more.
Read more →Which Drugs Target My Gene? A Free Drug Target Explorer
Paste a gene — or a whole gene set — and get every drug and clinical candidate acting on it: what each one treats, how far it reached in the clinic, and any CAR-T trials against it. Genes with an approved drug sort first, so a hit list ranks itself by how actionable it is. Disease and pathway views included. Free, no login.
Read more →Searching NCBI GEO Without Losing an Afternoon — Ranked Results & Direct Download Links
GEO's native search returns thousands of poorly-ranked, duplicate-heavy hits and buries download links in FTP folders. The free GEO Dataset Finder ranks datasets by relevance, groups SuperSeries clutter, and surfaces direct NCBI download links — including per-sample files. No login.
Read more →Design qPCR Primers in Seconds — Free, Exon-Junction-Aware, UCSC-Verified
Type a gene name, get validated qPCR primers instantly — Primer3-designed over RefSeq mRNA, exon-junction-spanning, with one-click UCSC In-Silico PCR verification. Free, human & mouse, now inside ChatGPT and Claude too.
Read more →Your junction-spanning primers show no In-Silico PCR hits — that's not a bug
A correctly designed exon-junction primer pair returns no contiguous hit in genomic DNA — the intron breaks the match — and people throw away good primers because of it. Why the genomic check cannot answer the question, and what to do if you work on an organism UCSC has no transcript target for.
Read more →Evaluations
1 postIndependent measurements of the tools and models this field runs on — method, numbers, and what went wrong.
Who Should Label a Dataset? Rules, a Decision Model and Two LLMs on the Same 300 Records
Four instruments on one question over 300 blinded GEO records: an expert-tuned rule system reached 43% accuracy, a calibrated decision model 68.8%, Claude Sonnet 5 69.1%, Claude Haiku 4.5 59.7%. On records two annotators agreed about, the models reached 84–92% and the rules 52.6%. The largest single effect was neither instrument nor model: rewording the option set moved inter-annotator kappa from 0.22 to 0.49.
Read more →Behind the Pipeline
1 postHow the briefings and tools are built, with the numbers: what was measured, what went wrong, what it cost.
Using Jev, a Decision Model, to Grade Biological Evidence at Scale
A regex filed "The immune cell atlas of human neuroblastoma" as a cell-line study. A calibrated decision model read the same text and said patient, p=1.00. What happened when we let it look at 26,000 records: agreement rates on FDA label polarity, GEO cohort type and trial eligibility, where each side was wrong, $0.03, and the rule we run.
Read more →