B Cell - Sequence-based

The B Cell - Sequence-based tool allows for the prediction of linear B-cell epitopes from protein sequences using several different methods. The tool includes BepiPred-3.0, which utilizes advanced protein language models, alongside several classical scale-based methods.

Methods

ML-based method

BepiPred-3.0 is currently the only ML-based method available.

Bepipred Linear Epitope Prediction 3.0
- Epitope probability from neural network; higher scores = more likely epitope.
- The BepiPred-3.0 server predicts both linear and discontinuous B-cell epitopes from protein sequence. It uses neural networks trained on protein language embeddings of epitope and non-epitope amino acids from crystal structures. The method leverages the ESM-2 protein language model (650M parameter version, esm2_t33_650M_UR50D) to generate numerical embeddings. These embeddings capture semantic and evolutionary information for each amino acid; the downstream neural network predicts epitope probability. A sequential smoothing (rolling mean) can be optionally used.
  [DOI: 10.1002/pro.4497]

Scale-based methods

Parameters such as hydrophilicity, flexibility, accessibility, turns, exposed surface, and polarity of polypeptide chains have been correlated with the location of continuous epitopes. All prediction calculations are based on scales for each of the 20 amino acids. Each scale consists of 20 values assigned to each amino acid residue based on their relative propensity to possess the property described by the scale.

The general method for a given residue i is to compute the score using amino acids in an interval of chosen length centered around residue i. For a window size n, the i − (n−1)/2 neighboring residues on each side of residue i are used. Unless specified, the score for residue i is the average of the scale values for these amino acids. In general, a window size of 5 to 7 is appropriate for finding potentially antigenic regions.

Chou & Fasman Beta-Turn Prediction
- Beta-turn propensity scale; higher scores = more turn likelihood. Turns are often found in epitope regions.
- The rationale for predicting turns to predict antibody epitopes is based on Pellequer et al., Immunology Letters, 36 (1993) 83-99. The Chou and Fasman scale is commonly used to predict beta turns.
- Reference: Chou PY, Fasman GD. 1978. Prediction of the secondary structure of proteins from their amino acid sequence. Adv Enzymol Relat Areas Mol Biol 47:45-148. [PMID: 364941]
- Scale: Amino acid propensity values for Chou & Fasman beta-turn scale.

A	C	D	E	F	G	H	I	K	L	M	N	P	Q	R	S	T	V	W	Y
0.66	1.19	1.46	0.74	0.6	1.56	0.95	0.47	1.01	0.59	0.6	1.56	1.52	0.98	0.95	1.43	0.96	0.5	0.96	1.14

Emini Surface Accessibility Prediction
- Surface accessibility; higher scores = more likely surface-exposed. S_n > 1.0 for hexapeptide indicates increased surface probability.
- The calculation uses a surface accessibility scale as a product (not a sum) within the window. The accessibility profile is obtained using the formula S_n = (n+4+i)(0.37)⁻⁶, where S_n is the surface probability and i varies from 1 to 6. A hexapeptide with S_n > 1.0 indicates increased probability of being found on the surface.
- Reference: Emini EA, Hughes JV, Perlow DS, Boger J. 1985. Induction of hepatitis A virus-neutralizing antibody by a virus-specific synthetic peptide. J Virol 55:836-839. [PMID: 2991600]
- Scale: Amino acid propensity values for Emini surface accessibility.

A	C	D	E	F	G	H	I	K	L	M	N	P	Q	R	S	T	V	W	Y
0.49	0.26	0.81	0.84	0.42	0.48	0.66	0.34	0.97	0.4	0.48	0.78	0.75	0.84	0.95	0.65	0.7	0.36	0.51	0.76

Karplus & Schulz Flexibility Prediction
- Flexibility (B-factor mobility); higher scores = more flexible regions, often antigenic.
- A flexibility scale was constructed from temperature B factors of the α-carbons of 31 proteins of known structure. The calculation is similar to the classical method. The center is the first amino acid of the six-amino-acid window; three scales describe flexibility instead of one.
- Reference: Karplus PA, Schulz GE. 1985. Prediction of chain flexibility in proteins. Naturwissenschaften 72:212-213. [Naturwissenschaften]

Kolaskar & Tongaonkar Antigenicity
- Semi-empirical antigenicity (~75% accuracy); higher scores = higher predicted antigenicity.
- A semi-empirical method using physicochemical properties and frequencies in known segmental epitopes. Application to many proteins has shown ~75% accuracy for predicting antigenic determinants.
- Reference: Kolaskar AS, Tongaonkar PC. 1990. A semi-empirical method for prediction of antigenic determinants on protein antigens. FEBS Lett 276:172-174. [PMID: 1702393]
- Scale: Amino acid propensity values for Kolaskar & Tongaonkar antigenicity.

A	C	D	E	F	G	H	I	K	L	M	N	P	Q	R	S	T	V	W	Y
1.064	1.412	0.866	0.851	1.091	0.874	1.105	1.152	0.93	1.25	0.826	0.776	1.064	1.015	0.873	1.012	0.909	1.383	0.893	1.161

Parker Hydrophilicity Prediction
- Hydrophilicity from HPLC retention; higher scores = more hydrophilic, surface-exposed, and antigenic.
- A hydrophilicity scale based on peptide retention times during HPLC on a reversed-phase column. A window of seven residues is used. The mean of the seven values is assigned to the fourth residue (i+3) in the segment.
- Reference: Parker JM, Guo D, Hodges RS. 1986. New hydrophilicity scale derived from high-performance liquid chromatography peptide retention data: correlation of predicted surface residues with antigenicity and X-ray-derived accessible sites. Biochemistry 25:5425-5432. [PMID: 2430611]
- Scale: Amino acid propensity values for Parker hydrophilicity.

A	C	D	E	F	G	H	I	K	L	M	N	P	Q	R	S	T	V	W	Y
2.1	1.4	10.0	7.8	−9.2	5.7	2.1	−8.0	5.7	−9.2	−4.2	7.0	2.1	6.0	4.2	6.5	5.2	−3.7	−10.0	−1.9

Input Data

B-Cell Sequence Input

Sequence Text: Enter protein sequences directly into the input field. This supports:
- FASTA Format: Standard text-based format (e.g., >header\nSEQUENCE).
- Plain Text: Raw amino acid sequences (e.g., MKAIL...). * UniProt IDs: To use a UniProt ID:
- Check the UniProt ID box.
- Enter the accession ID (e.g., P02185) to automatically retrieve the sequence.

Note

Due to the scales for each of the 20 amino acids, the tool does not accept ambiguity codes such as X, B, Z, and J.
File uploading is currently not supported in the UI.

Parameter Selection

Method Selection

Select the desired prediction method from the dropdown menu. Bepipred Linear Epitope Prediction 3.0 is the latest method.

Method-Specific Parameters

Bepipred 3.0:
- Window Size: The number of amino acids considered in the sliding window.
- Prediction Scoring:
  - Majority Vote: Uses an ensemble of models to vote on the prediction.
  - Variable Threshold: Uses a specific threshold value for prediction.
- Include sequence length in ESM encodings: Adds sequence length as a feature to the ESM-2 embeddings (equivalent to -add_seq_len in CLI).
- Threshold: (If Variable Threshold is selected) Adjust the sensitivity/specificity of the prediction. Higher thresholds result in fewer but more confident predictions.
  - Default: 0.1512 (Corresponds to specificity > 0.85)
scale-based methods (Chou-Fasman, Emini, Karplus & Schulz, Kolaskar-Tongaonkar, Parker):
- Window Size: The number of amino acids considered in the sliding window.
  - Default: 7 (except Emini which defaults to 6)
- Center Position: (Implicit in window size) The score is assigned to the central residue of the window.

Window Size Defaults and Ranges

Method	Default window size	Window size range
Emini	6	6
Parker	7	1-25
Karplus & Schulz	7	7
Kolaskar & Tongaonkar	7	1-25
Chou & Fasman	7	1-25
BepiPred-3.0	9	1-50

Results

Tabular Results

The following tables are available:

Residue Table: A detailed view of scores for each individual residue (center position of the window) in the sequence. Columns can be shown or hidden via “Display Columns”. Columns are grouped by method to mirror the Display Columns dialog.

Linear Epitope Table: Summarizes contiguous stretches of predicted epitopes.

Note

Only regions with at least 5 consecutive epitope residues are shown in the linear epitope table.

Sequence Table: One row per input sequence with aggregate statistics (e.g., sequence length, number of predicted epitopes). Available columns depend on the selected methods; use the app’s Display Columns to show or hide metrics.

Bcell Sequence Plots

Score profiles along the sequence; each method has its own plot.

Example: Chou & Fasman prediction plot

Chou & Fasman prediction plot example

Axes: X-axis = residue position; Y-axis = score averaged in the specified window for each residue.

Elements: Score line; red threshold line; green shading (positive prediction); pink shading (negative prediction).

Interpreting Scores

For all methods, higher scores indicate regions more likely to be epitopes (or epitope-correlated properties). BepiPred-3.0 and Kolaskar & Tongaonkar directly predict epitope likelihood or antigenicity; Chou & Fasman, Emini, Karplus & Schulz, and Parker predict structural or physicochemical properties (beta-turns, surface accessibility, flexibility, hydrophilicity) that correlate with epitope location.

Note

The scale-based methods do not predict epitopes per se—they guide researchers to regions worth further exploration.

Table

The Residue Table displays per-residue scores and assignments.

Score columns (e.g., bepipred score, chou & fasman score): Use these with percentile ranks to compare residues.
Assignment columns (E or n): Epitope vs non-epitope at the selected threshold.

Interpretation: Higher scores correspond to residues more likely to be part of an epitope.

Graph

The Bcell Sequence Plots tab shows score profiles along the sequence.

Example: Chou & Fasman prediction plot

Chou & Fasman prediction plot example

See Bcell Sequence Plots above for axes and elements (score line, threshold, green/pink shading). Higher peaks indicate stronger propensity; use the plot to identify contiguous high-scoring regions that may warrant further investigation.

Per-method details

BepiPred-3.0

Interpretation: Higher scores = greater likelihood of epitope.

Chou & Fasman

Interpretation: Higher scores = greater turn propensity. Beta turns are often found in antigenic regions (Pellequer et al.).

Emini

Interpretation: Higher scores = more likely surface-exposed.

Karplus & Schulz

Interpretation: Higher scores = more flexible regions, often antigenic.

Kolaskar & Tongaonkar

Interpretation: Higher scores = higher predicted antigenicity. Based on physicochemical properties and frequencies in known epitopes.

Parker

Interpretation: Higher scores = more hydrophilic, more likely surface-exposed and antigenic.