---
title: ANARCI on the platform
description: >-
  Number antibody sequences to canonical schemes and annotate CDRs, framework
  regions, and germline genes with ANARCI.
tags:
  - anarci
  - antibodies
---

Before you can compare an antibody to another, humanize it, or point at a specific CDR loop, you need every residue to have a canonical number — position 27 in one antibody has to mean the same structural place as position 27 in the next. ANARCI assigns that numbering from sequence alone, in whichever standard scheme your workflow expects (IMGT, Kabat, Chothia, AHo, or Wolfguy), and in doing so carves the sequence into its CDR and framework regions and names the closest germline genes. It is the entry step for most of the [antibody engineering](/docs/antibodies) arc.

> **Known accuracy issue**

> ANARCI can return CDR and framework region boundaries that don't match the requested numbering scheme — framework residues labelled as CDR residues — while the job still reports status `SUCCESS`. Verify CDR boundaries against an independent numbering tool before relying on them for humanization or paratope work.

## Try it

**Try this prompt** — [open in a guest chat](/chat/new?guest=true&prompt=Run%20IgBLAST%20on%20the%20adalimumab%20heavy%20chain%20to%20identify%20V%2FD%2FJ%20germline%20genes%20and%20CDR%20boundaries%2C%20then%20score%20joint%20VH%2FVL%20humanness%20with%20Humatch%20%E2%80%94%20does%20the%20fully%20human%20antibody%20score%20above%20therapeutic%20threshold%2C%20and%20recommend%20whether%20the%20V-gene%20assignment%20suggests%20a%20manufacturer-specific%20humanization%20liability%20for%20a%20biosimilar%20program%3F).

> Run IgBLAST on the adalimumab heavy chain to identify V/D/J germline genes and CDR boundaries, then score joint VH/VL humanness with Humatch — does the fully human antibody score above therapeutic threshold, and recommend whether the V-gene assignment suggests a manufacturer-specific humanization liability for a biosimilar program?

_ANARCI and IgBLAST are the numbering/annotation pair — either can seed a humanization workflow._

## What ANARCI does

ANARCI (Antigen receptor Numbering And Receptor ClassIfication) takes one or more antibody variable-region sequences and assigns each residue a canonical position number. From that numbering it derives the CDR and framework boundaries and, when asked, aligns the sequence to germline V and J genes to name the closest match. The result is the coordinate system the rest of the antibody cluster reads: humanness scorers, developability profilers, and paratope predictors all reason in terms of numbered positions.

## Citations

> Dunbar J, Deane CM. ANARCI: antigen receptor numbering and receptor classification. *Bioinformatics*. 2016;32(2):298-300. doi: [10.1093/bioinformatics/btv552](https://doi.org/10.1093/bioinformatics/btv552)

## When to reach for it

- **Canonical numbering** — you need residues numbered in IMGT, Kabat, Chothia, AHo, or Wolfguy so positions are comparable across antibodies.
- **CDR identification** — you want the six CDR loops and four framework regions carved out of a raw VH or VL sequence.
- **Germline assignment** — you want the closest germline V/J gene named for each chain.

Reach for [IgBLAST](/docs/igblast) instead when you specifically need V(D)J germline gene *hits* with alignment statistics and somatic-hypermutation counts. Reach for [BioPhi](/docs/biophi) or [Humatch](/docs/humatch) when the question is humanness rather than numbering. ANARCI chains forward into both `biophi_humanize` and `antibody_developability`.

## How it fits the workflow

ANARCI is a numbering step, not a scoring step — its output is the substrate other tools consume. A common arc is `anarci_number → biophi_humanize → antibody_developability`: number the sequence, humanize it, then check that the humanized candidate is manufacturable. Because it accepts a `therapeutic_name`, you can also start from a known therapeutic and let the platform resolve the VH/VL from Thera-SAbDab before numbering.

## How it runs

ANARCI runs as an asynchronous CPU job on Modal compute — you submit one or more sequences, the platform tracks the job and polls it, and a result card lands on the canvas when it completes, typically within two to three minutes.

## Read next