You provide a protein as an amino acid sequence — pasted directly, fetched from a UniProt accession, or uploaded in FASTA. The one thing that most affects your result is submitting the true expressed construct: use standard residues only, and match the mature sequence (signal peptides removed, tags included) to the protein you actually plan to express.

Supported inputs

DeepSol accepts protein sequences for E. coli expression solubility prediction. The CNN architecture exploits k-mer structure and combines sequence-level features with structural properties from SCRATCH-1D predictions.

Webapp input methods

The .bloom* webapp provides two input modes:

Sequence input (default)

Protein Sequence: Paste the amino acid sequence in single-letter code directly
Format: Raw sequence without FASTA header
Example: MVLSPADKTNVKAAWGKVGAHAGEYGAEALERMFLSFPTTKTYFPHFDLSHGSAQVKGHGKKVADALTNAVAHVDDMPNALSALSDLHAHKLRVDPVNFKLLSHCLLVTLAAHLPAEFTPAVHASLDKFLASVSTVLTSKYR

UniProt Accession input

Accession ID: Enter a valid UniProt accession (e.g., P68871)
Auto-fetch: The system automatically retrieves the protein sequence from UniProt
Validation: Real-time checking of accession format

File-based inputs

FASTA files (preferred): standard FASTA format with header lines starting with >; UTF-8 encoded.
Raw sequences: single amino acid sequences in one-letter code; one sequence per line.
Batch CSV: header sequence; optional id column is echoed back in results.

Sequence requirements

Amino acids: Standard 20 amino acids only (A, C, D, E, F, G, H, I, K, L, M, N, P, Q, R, S, T, V, W, Y).
Sequence length: Recommended range 50-2000 residues; very short or very long sequences may have reduced accuracy.
No ambiguous residues: Avoid X, B, Z, J, or other non-standard characters.
Single chain: Submit one polypeptide chain per entry; multi-chain complexes should be split.

Webapp parameters

The following options are available in the .bloom* webapp interface:

SCRATCH-1D features

Use SCRATCH-1D (toggle, default: enabled)
Enables additional structural feature extraction using SCRATCH-1D predictions

Model selection

Model: Choose between deepsol1, deepsol2, or deepsol3
Each version has different training data and performance characteristics

Processing mode

Mode: Select the operation type
- preprocess - Data preprocessing only
- train - Model training (advanced)
- decode - Generate predictions
- cv - Cross-validation mode

Pipeline stage

Stage: Select processing stage (1 or 2)

Compute device

Device: Choose between cpu or cuda (GPU acceleration)

Advanced options (optional)

Config File Path: Path to custom DeepSol configuration
Keras Backend: Specify Keras backend (tensorflow, theano, cntk)
CUDA Root Path: Root path for CUDA installation if using GPU
Test File Name: Name of test data file for evaluation

Validation rules

Sequences containing non-standard amino acids will be rejected.
Sequences shorter than 10 residues may produce unreliable predictions.
Very long sequences (>3000 residues) may timeout; consider domain-based splitting.
Numerical characters, spaces, and special symbols are stripped automatically.

Preparing data

Remove signal peptides and pro-domains if predicting mature protein solubility.
Ensure sequences represent the actual expressed construct (include tags if used).
For fusion proteins, submit the complete fusion sequence.
Remove any formatting characters (line breaks within sequences, numbering).

Submission tips

Keep batches under ~1,000 sequences for faster feedback; large collections should be chunked.
Retry on 429 with exponential backoff; respect workspace rate limits.
Store original input order externally; outputs are returned in processed order.
Include descriptive FASTA headers to track sequences through the pipeline.

DeepSol inputs and parameters