Skip to main content

Table 2 Per-segment performance for TMH (transmembrane helices). *

From: TMbed: transmembrane proteins predicted through language model embeddings

  TMH (571/2936)
Recall (%) Precision (%) Qok (%) Qnum (%) Qtop (%)
TMbed 88.7 ± 0.6 88.7 ± 0.7 62.4 ± 3.7 86.0 ± 2.3 96.4 ± 2.7
DeepTMHMM 80.0 ± 2.4 80.5 ± 2.4 46.2 ± 4.8 85.7 ± 3.5 96.3 ± 2.2
TMSEG 74.5 ± 2.4 77.1 ± 1.7 35.6 ± 2.4 69.9 ± 2.7 83.8 ± 4.7
TOPCONS2 76.4 ± 1.5 78.4 ± 0.8 41.0 ± 3.1 74.4 ± 3.3 91.7 ± 3.1
OCTOPUS 71.6 ± 1.5 75.7 ± 1.4 36.0 ± 2.8 67.6 ± 3.4 87.5 ± 3.1
Philius 70.8 ± 2.2 73.7 ± 0.8 34.2 ± 3.7 66.9 ± 3.4 87.5 ± 2.9
PolyPhobius 76.0 ± 2.1 76.4 ± 1.1 40.3 ± 3.5 74.5 ± 2.8 86.8 ± 2.7
SPOCTOPUS 71.5 ± 1.2 75.8 ± 1.2 35.7 ± 3.3 67.4 ± 5.5 87.2 ± 3.4
SCAMPI2 (MSA) 72.3 ± 2.7 74.1 ± 1.5 33.5 ± 3.0 72.2 ± 4.5 90.6 ± 3.5
CCTOP1 77.0 ± 1.7 79.4 ± 1.0 41.9 ± 3.6 82.6 ± 2.7 92.6 ± 2.6
HMM-TM (MSA)2 73.3 ± 1.7 72.5 ± 1.2 33.5 ± 1.4 72.1 ± 3.0 88.3 ± 4.2
  1. *Segment performance for transmembrane helix (TMH) prediction based on 571 alpha helical TMPs (α-TMP) with a total of 2936 TMHs. Recall, Precision, Qok, Qnum, and Qtop were averaged over the five independent cross-validation test sets; error margins given for the 95% confidence interval (1.96*standard error); bold: best values for each column; italics: differences statistically significant with over 95% confidence (only computed between best and 2nd best).
  2. 1Evaluation missing for one of 571 α-TMPs.
  3. 2Evaluation includes only 552 of the 571 α-TMPs due to runtime errors of the method.