Models

Dorado basecalling relies upon machine learning models to decode the raw nanopore sequencing data. The appropriate model for your data will be automatically selected by Dorado basecaller using the model selection complex. However, you can also manually select a model using the naming conventions below.

There are a number of factors which define a basecalling model, but the key factors are broadly summarised by the balance of performance and accuracy that models provide, and the data that the model was trained to accurately decode.

Understanding model names

The names of Dorado models are systematically structured, each segment corresponding to a different aspect of the model, which include both chemistry and run settings (defined here). Below are some examples of simplex basecalling models:

dna_r10.4.1_e8.2_400bps_sup@v5.2.0
rna004_130bps_hac@v5.2.0

{analyte}_{pore}_{chemistry}_{speed}@version

Sequencing condition

Models are trained on carefully curated datasets for specific nanopore sequencing condition and as such they are each assigned specific names to denote which condition they are paired.

The sequencing condition will typically denote the following features:

Analyte Type - dna / rna004

This denotes the type of analyte being sequenced. For DNA sequencing, this will be dna. If you are using a Direct RNA Sequencing Kit, this will be rna004.

Pore Type - r10.4.1

This section corresponds to the type of flow cell used. For instance, FLO-MIN114 / FLO-FLG114 is indicated by r10.4.1.

Chemistry Type - e8.2

This represents the chemistry type, which corresponds to the kit used for sequencing. For example, Kit 14 chemistry is denoted by e8.2.

Translocation Speed - 130bps / 260bps / 400bps:

This parameter, defines the speed of translocation.

Speed and Accuracy

Typically for each model generation, 3 models are available and are named fast, hac (high-accuracy), and sup (super-accurate). These are in order of increasing basecalling accuracy where fast is the least accurate and sup is the most accurate. In general, larger models are more accurate but are more computationally expensive to evaluate.

As such, we recommend the hac model for most users as it strikes the best balance between accuracy and computational cost.

Model Version Numbers

Tip

We recommend that users use the latest models for the best results.

Basecalling models are frequently updated to improve accuracy and performance. The model version is identified using the following form v{major}.{minor}.{patch} for example v4.3.0.

Simplex basecalling models are identified by only one version but modification model names contains two version numbers for example: dna_r10.4.1_e8.2_400bps_hac@v4.3.0_6mA@v1 and follow the format: {simplex_model@version}_{modification@version}

This is because all modification models are paired with a specific simplex model. The first version identifies the simplex model while the second identifies the version of the modification model. As such, modification model version numbers are reset on each simplex model update.

For example, a 6mA@v1 modification model compatible with the v4.3.0 simplex model is more recent than a 6mA@v2 modification model compatible with av4.2.0 simplex model.