AAV Capsid Production Prediction User Guide

AAV ViaPredictor is an AI-based AAV yield prediction tool jointly developed by Cyagen Biosciences and the AI Innovation Center of Tsinghua Panyu Research Institute. It integrates multi-dimensional feature data with deep learning frameworks to predict yields across different serotype vectors, supporting yield optimization for gene therapy research and production.

The tool accepts nucleotide sequences for AAV2 and AAV9 serotypes to perform yield predictions. Specific application conditions are shown below.

Figure 1. Operation diagram. The input fragment must be a nucleotide sequence.

• Supported Variants: Insertions of 21–30 nucleotides (7–10 amino acids) between wild-type positions 587 and 588.
• Reference Sequence: Wild-type positions 561–592 DEEEIRTTNPVATEQYGSVSTNLQRGN^RQAAT (^ marks the 587–588 insertion site).
• Insert Structure: An N-terminal TG and a C-terminal GLS restriction site are added automatically, forming TG~GLS where ~ is your input fragment.

• Substitution: Amino acid substitution at wild-type positions 586–588 (SAQ); the SAQ codons are AGTGCCCAA.
• Insertion: 21 nucleotides (7 amino acids) inserted between positions 588 and 589.
• Reference Sequence: Wild-type positions 562–593 NEEEIKTTNPVATESYGQVATNHQSAQ^AQAQT (^ marks the 588–589 insertion site).
• Requirement: The inserted fragment must be exactly 7 amino acids long.

Note: Input nucleotide sequences, not amino acid sequences.

1.Choose a serotype. For AAV2, enter the insert fragment. For AAV9, enter the 9 nt substitution segment followed by the 21 nt insert, totaling 30 nt.

2.Paste FASTA sequences or one nucleotide sequence per line. You can also load examples, select a .fasta / .fa file, or format the input as FASTA.

3.Check each sequence in the preview. A green check means valid; red text shows the reason for an issue. Checks update as you edit.

4.Click Predict to see the task status, result table and yield chart below the input. After completion, sort the table or download all results as CSV.

5.Tasks submitted while signed in can be viewed and deleted in History. View restores the serotype and sequences and displays the task results on the same page.

The result is a table: the name identifies each input sequence, and the predicted score is the model output (log2-scaled). A higher score means higher yield.

Up to 20 sequences per run.

Runs made while signed in appear in History and can be reviewed or deleted. Runs made without signing in are not listed in anyone's history.

The record and its sequences are permanently removed.