ProteinScout — sp|P02945|BACR_HALSA Bacteriorhodopsin OS=Halobacterium salinarum (strain ATCC 700922 / JCM 11081 / NRC-1) OX=64091 GN=bop PE=1 SV=2

sp|P02945|BACR_HALSA Bacteriorhodopsin OS=Halobacterium salinarum (strain ATCC 700922 / JCM 11081 / NRC-1) OX=64091 GN=bop PE=1 SV=2

ML layer vs single-scale baseline

Held-out ROC-AUC on real labeled benchmarks (WALTZ-DB amyloid, DisProt disorder, UniProt TRANSMEM). The ML model uses windowed multi-scale + composition features; the baseline is the single published scale the classic method uses. Δ > 0 means the ML layer adds value.

task	baseline scale	baseline AUC	ML AUC	Δ AUC	ML AUPRC
amyloid	AGGRESCAN a3v (mean)	0.811	0.849	+0.039	0.724
disorder	TOP-IDP (windowed mean)	0.775	0.815	+0.040	0.561
tm	Kyte-Doolittle (window-19 mean)	0.951	0.964	+0.013	0.737

track	precision	recall	F1
TM (scale)	0.82	0.65	0.73
TM (ML)	0.83	0.91	0.87

Biophysical interpretation

Bacteriorhodopsin from *Halobacterium salinarum* is a well-folded integral membrane protein with a strongly negative charge-hydropathy margin (-0.438), consistent with its known structure as a seven-transmembrane helix protein. The predicted topology shows 6 TM helices by the hydrophobicity scale (residues 28-41, 60-79, 103-120, 125-138, 154-168, 193-200) with N-terminus facing outward, though the canonical seventh helix may be missed due to threshold settings. Both scale and ML methods identify extensive aggregation-prone regions (23-46, 54-84, 103-115) that overlap significantly with the transmembrane segments, which is expected for membrane proteins where hydrophobic residues necessary for membrane insertion can appear as false-positive aggregation signals. The absence of any predicted disordered regions by both methods aligns with bacteriorhodopsin's known compact, helical structure, and the lack of signal peptide is consistent with its insertion via the TAT pathway in halophilic archaea.

Generated by Claude (Anthropic) from the computed profiles above.

Predicted regions

Transmembrane helices — published scale

Kyte-Doolittle window-15 hydropathy peak ≥ 1.0 over ≥7 residues.

start	end	len	peak hydropathy	mean
28	41	14	1.72	1.35
60	79	20	1.98	1.42
103	120	18	2.18	1.36
125	138	14	1.67	1.43
154	168	15	2.11	1.63
193	200	8	1.91	1.34
214	231	18	1.86	1.50

Transmembrane helices — ML model (UniProt-trained)

Logistic model on windowed multi-scale features; P(TM) ≥ 0.5 over ≥7 residues.

start	end	len	peak P(TM)
24	42	19	0.99
58	79	22	0.99
97	110	14	0.98
127	169	43	1.00
186	207	22	0.98
209	237	29	0.99

TM topology — positive-inside rule

N-terminus predicted out (von Heijne 1992; inside loops K+R density 0.099).

start	end	orientation
28	41	out->in
60	79	in->out
103	120	out->in
125	138	in->out
154	168	out->in
193	200	in->out
214	231	out->in

Aggregation-prone regions — published scale

AGGRESCAN a3v hot-spots (≥5 residues above threshold).

start	end	len	peak a3v	sequence
1	7	7	0.73	MLELLPT
23	46	24	1.45	WIWLALGTALMGLGTLYFLVKGMG
54	84	31	1.01	KFYAITTLVPAIAFTMYLSMLLGYGLTMVPF
90	95	6	0.73	PIYWAR
98	114	17	1.14	DWLFTTPLLLLDLALLV
120	126	7	1.23	TILALVG
129	139	11	0.77	GIMIGTGLVGA
141	169	29	1.53	TKVYSYRFVWWAISTAAMLYILYVLFFGF
182	186	5	0.73	STFKV
188	205	18	1.49	RNVTVVLWSAYPVVWLIG
210	226	17	1.40	GIVPLNIETLLFMVLDV
230	238	9	1.16	VGFGLILLR

Aggregation-prone regions — ML model (WALTZ-DB-trained)

Hexapeptide amyloid model scanned across the sequence; P(amyloid) ≥ 0.5.

start	end	len	peak P(amyloid)
4	9	6	0.54
23	45	23	0.95
54	84	31	0.92
96	101	6	0.59
103	115	13	0.75
119	145	27	0.91
148	171	24	0.96
182	207	26	0.95
211	228	18	0.92
230	240	11	0.95

Intrinsically disordered regions — published scale

None predicted.

Intrinsically disordered regions — ML model (DisProt-trained)

None predicted.

MoRF / binding-region hints

None predicted.

Coiled-coil regions

None predicted.

Low-complexity regions

None predicted.

aa	%
L	14.9%
A	11.5%
G	9.9%
V	8.8%
T	7.3%
I	5.7%
S	5.3%
F	5.0%
P	4.6%
E	4.2%
Y	4.2%
D	3.8%
M	3.8%
W	3.1%
K	2.7%
R	2.7%
Q	1.5%
N	1.1%

Sequence

<span class='pos'>    1</span> MLELLPTAVEGVSQAQITGRPEWIWLALGTALMGLGTLYFLVKGMGVSDPDAKKFYAITT
<span class='pos'>   61</span> LVPAIAFTMYLSMLLGYGLTMVPFGGEQNPIYWARYADWLFTTPLLLLDLALLVDADQGT
<span class='pos'>  121</span> ILALVGADGIMIGTGLVGALTKVYSYRFVWWAISTAAMLYILYVLFFGFTSKAESMRPEV
<span class='pos'>  181</span> ASTFKVLRNVTVVLWSAYPVVWLIGSEGAGIVPLNIETLLFMVLDVSAKVGFGLILLRSR
<span class='pos'>  241</span> AIFGEAEAPEPSAGDGAAATSD

ProteinScout report

Per-residue tracks (blue/green = published scale, orange = ML)

ML layer vs single-scale baseline

Validation vs UniProt annotation

Whole-protein disorder — charge-hydropathy plot

Biophysical interpretation

Predicted regions

Transmembrane helices — published scale

Transmembrane helices — ML model (UniProt-trained)

TM topology — positive-inside rule

Aggregation-prone regions — published scale

Aggregation-prone regions — ML model (WALTZ-DB-trained)

Intrinsically disordered regions — published scale

Intrinsically disordered regions — ML model (DisProt-trained)

MoRF / binding-region hints

Coiled-coil regions

Low-complexity regions

Amino-acid composition

Sequence