DEV Community: devin2xing

FASTA vs FASTQ — Format Differences, Examples and Conversion

devin2xing — Thu, 12 Nov 2026 08:04:01 +0000

FASTA and FASTQ are the two workhorse text formats of bioinformatics. FASTA is older and simpler;
FASTQ was introduced with next-generation sequencing to carry a per-base confidence score
alongside each base.

FASTA

A FASTA record is a header line beginning with > followed by one or more lines of
sequence:

>seq1 sample description
ATGGCCATTGTAATGGGCCGCTGAAAGGGTGCCCGATAG
>seq2
ATGGCCTTTGTAATGGGCCGCTGAAAGGGTGCCCGATAG

Sequence lines are conventionally wrapped at 60 or 70 characters, but any width is legal. A file
can hold any number of records.

What FASTA is used for

Reference genomes and transcriptomes
Protein sequences
BLAST databases
Multiple sequence alignment input
Anywhere the sequence itself is the payload and per-base confidence is irrelevant

FASTQ

A FASTQ record has four lines:

@read1
ATGGCCATTGTAATGGGCCGCTGAAAGGGTGCCCGATAG
+
IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII

@ header — same content as a FASTA header, different marker
Sequence
+ separator, optionally repeating the header
Quality string — exactly one character per base

What the quality string means

Each character encodes a Phred quality score Q, which is a log-scale estimate of the probability
that the base is wrong:

Q = -10 × log10(P_error)

The character is the score plus an offset. Modern Illumina data uses Phred+33 (Sanger encoding),
so:

Character	ASCII	Phred Q	Error probability
`!`	33	0	1 in 1
`+`	43	10	1 in 10
`5`	53	20	1 in 100
`?`	63	30	1 in 1,000
`I`	73	40	1 in 10,000

Older Illumina pipelines used Phred+64, which produces characters like h for a
decent score. Mixing the two encodings silently corrupts quality interpretation — one of the
reasons FASTQ files carry a "encoding" note in their metadata.

Converting FASTQ to FASTA

This throws away the quality string. It is a one-line transformation per record, and it is
completely safe — but it is also irreversible. Once the quality data is gone it
cannot be reconstructed.

Convert when:

Your downstream tool only accepts FASTA (BLAST, most aligners' reference input)
You are assembling or annotating and quality has already been used for filtering
You need a compact archive of the sequences

Do not convert when you still need to trim, filter or call variants — those steps depend
on the quality scores.

Converting FASTA to FASTQ

Going the other way is synthesising information that never existed. You cannot invent real
quality scores, so any converter has to write a placeholder. Ours writes the highest quality
character throughout, which means "assume every base is correct".

This is legitimate when an old tool insists on FASTQ input but you genuinely have no quality
data — for example a sequence reconstructed from a paper. It is not legitimate for
feeding an aligner that will use those scores to decide which mismatches to trust.

Quick reference

	FASTA	FASTQ
Header marker	`>`	`@`
Lines per record	2+ (wrapped)	exactly 4
Quality scores	no	yes, one per base
Typical size	1× sequence	~2× sequence
Typical source	reference, protein	sequencer output

Try it

Convert in either direction with FASTA → FASTQ or
FASTQ → FASTA, or compute GC content and sequence statistics
with the GC content calculator. All of them run locally in Python
on this server — nothing is sent to a third party.

Convert in either direction with FASTA to FASTQ or FASTQ to FASTA, and get sequence statistics from the GC content calculator.

Lipinski Rule of Five Explained — With Worked Examples

devin2xing — Thu, 12 Nov 2026 07:58:01 +0000

The rule of five was published by Christopher Lipinski and colleagues in 1997, based on an
analysis of compounds that made it into clinical trials versus those that failed. It is a set of
four thresholds, each involving the number five or a multiple of it — hence the name.

The four rules

Property	Threshold	What it reflects
Molecular weight	≤ 500 Da	Size — large molecules diffuse poorly
logP (octanol/water)	≤ 5	Lipophilicity — too greasy means poor solubility
Hydrogen bond donors (HBD)	≤ 5	Count of N–H and O–H groups
Hydrogen bond acceptors (HBA)	≤ 10	Count of N and O atoms

The usual interpretation is "no more than one violation". Two or more violations is where the
probability of poor oral absorption rises sharply.

Where the numbers come from — and what they really mean

The thresholds were not derived from first principles. They are the 90th percentile of the
distribution for compounds that reached phase II trials. In other words: the rule describes the
space that historically successful oral drugs occupied, not a physical law.

That has two consequences worth remembering:

Passing the rule does not make a compound a drug. It makes it statistically similar to ones that entered trials.
Failing the rule does not make a compound useless. Many antibiotics, antifungals, PROTACs and natural products are far outside the space and work very well.

Worked example: aspirin

SMILES: CC(=O)Oc1ccccc1C(=O)O

Property	Value	Pass?
MW	180.16	✅
logP	1.19	✅
HBD	1 (the carboxylic acid OH)	✅
HBA	4 (three O, one carbonyl O counted separately)	✅

Aspirin passes on all four. Run the same input through the drug-likeness
checker and you get the same table in one request.

The Veber extensions

Lipinski's rule says nothing about flexibility, which turns out to matter for oral
bioavailability. Veber et al. (2002) added two criteria that are now usually reported alongside:

Rotatable bonds ≤ 10 — fewer conformations, better absorption
Polar surface area (TPSA) ≤ 140 Å² — beyond this, passive permeability drops

A molecule can pass Lipinski and fail Veber — a long flexible chain with one polar group, for
example. Reporting both gives a much more honest picture, which is why our
checker shows all six numbers.

Common mistakes

Counting hydrogen bond acceptors wrong

Lipinski's HBA count is the sum of nitrogen and oxygen atoms, not the number of lone pairs and
not the number of acceptor groups. A nitro group counts as three, a carboxylate as two.

Using logP when you mean logD

The rule uses logP — the partition coefficient of the neutral form. At physiological pH, an
ionisable acid or base has a very different effective lipophilicity (logD). A carboxylic acid
with logP 4.5 may have logD near 1 at pH 7.4. If you are reasoning about absorption, logD is
usually the more relevant number.

Applying it to biologics

The rule was derived from small molecules. Applying MW ≤ 500 to an antibody or a peptide is
meaningless — those molecules are designed to work by entirely different mechanisms.

Use it as a filter, not a verdict

The most useful application is triage: run a large library through the rule, flag compounds with
two or more violations, and spend your synthesis budget elsewhere first. Keep the flag, not the
decision. Run the checker on any SMILES string, or paste a whole library
into the batch converter to compute properties in bulk.

Run the rule on any SMILES string with the free drug-likeness checker, or paste a whole library into the batch converter.

SMILES vs InChI vs InChIKey — Differences Explained (2026)

devin2xing — Thu, 12 Nov 2026 07:58:00 +0000

If you have ever tried to match compounds across two databases and found that the same molecule
appears twice, the culprit is usually the representation format. SMILES, InChI and InChIKey all
describe a structure, but they differ in what they guarantee — and only one of them is safe as a
primary key.

SMILES: compact, human-writable, not canonical

SMILES (Simplified Molecular Input Line Entry System) writes a molecule as a traversal of its
graph. Atoms are element symbols, bonds are implicit or written as -, =,
#, branches use parentheses and rings use digit closures.

Aspirin, for example, can be written as:

CC(=O)Oc1ccccc1C(=O)O
OC(=O)c1ccccc1OC(C)=O
O=C(O)c1ccccc1OC(=O)C

All three are valid and describe the same molecule. That flexibility is exactly the problem: a
string comparison of two SMILES is not a comparison of two molecules.

What SMILES is good for

Pasting into a structure editor or drawing tool
Compact storage inside a script or a database column you control
SMARTS substructure queries (SMARTS is SMILES extended with wildcards)
Input to almost every cheminformatics toolkit

What SMILES is bad for

Deduplication — unless you canonicalise first
Cross-database lookup — different toolkits canonicalise differently
Regulatory or archival identifiers

Most toolkits do have a canonical SMILES algorithm, but "canonical" is toolkit-specific. RDKit's
canonical SMILES and OpenBabel's canonical SMILES are both deterministic and both perfectly
valid — and they are not guaranteed to produce the same string. That makes canonical SMILES a
good internal key and a poor external one.

InChI: canonical by standard

InChI (International Chemical Identifier) is maintained by IUPAC and designed specifically to be
a canonical, non-proprietary identifier. Every conforming implementation must produce the same
string for the same structure. The format is layered:

InChI=1S/C9H8O4/c1-6(10)13-8-5-3-2-4-7(8)9(11)12/h2-5H,1H3,(H,11,12)

1S — version 1, standard InChI
C9H8O4 — molecular formula layer
c1-6(10)13-... — connectivity layer
h2-5H,1H3,(H,11,12) — hydrogen layer

Separate layers also cover charge, stereochemistry (including /t and /m
sub-layers) and isotopes. If a structure has no stereo, those layers are simply absent.

Why InChI exists

Before InChI, every vendor had its own canonical form. InChI gave the community a single
reference string that could be published in a paper or deposited with a patent and re-derived
later by anyone. PubChem, ChEMBL, Wikidata and most patent databases index by InChI and
InChIKey.

The downside: length

A full InChI is long, contains punctuation that needs escaping in URLs, and grows with molecule
complexity. For a small molecule it is ~60 characters; for a peptide or a natural product it can
be hundreds. Indexing millions of them is possible but not free.

InChIKey: fixed-width hash

The InChIKey is a hash of the InChI, always exactly 27 characters:

BSYNRYMUTXBXSQ-UHFFFAOYSA-N

The three blocks are meaningful:

First 14 — the molecular skeleton (connectivity plus formula)
Next 8 — stereochemistry, charge and isotope layers
Final 1 — a check character to catch transcription errors

Because the first block ignores stereo, two enantiomers share the same 14-character prefix and
differ only in the second block. That is convenient for "find me anything with this skeleton"
queries.

What InChIKey cannot do

It is a one-way hash. You cannot recover a structure from a key, and you cannot convert a key
into anything else. Keep the InChI or the original structure file alongside it. Also note that
the key is case-sensitive — BSYNRYMUTXBXSQ-UHFFFAOYSA-N and its lowercase form are
different strings to a database.

Which one should you use?

Task	Use
Pasting into a drawing tool	SMILES
Deduplicating a library	InChI (or InChIKey)
Primary key in a database	InChIKey, with InChI stored alongside
Matching against PubChem / ChEMBL	InChIKey
SMARTS substructure search	SMILES / SMARTS
Filing a patent or a regulatory dossier	InChI

Converting between them

All three conversions are one click on this site: SMILES to InChI,
InChI to SMILES, SMILES to
InChIKey and InChI to InChIKey. Two rules of thumb:

Always convert toward InChI when you cross a system boundary, never away from it.
Never rely on a SMILES string being stable across tools; canonicalise at the point of entry.

SeqChem has free browser tools for all of this: SMILES to InChI, InChI to SMILES and SMILES to InChIKey.