Table 6 Prediction error of the RF, ULSTM, BLSTM and CNN-LSTM models, as measured by RMSD metric in Å units. The three extracellular loops (ECL) of the APO, FA, PIA simulations for 2rh1 and 3p0g states.

From: GPCR molecular dynamics forecasting using recurrent neural networks

Region

MD-RMSD

RF

ULSTM

BLSTM

CNN-LSTM

APO

2rh1

ECL1\(^{**+}\)

2.0800

0.2440\(^{**-}\) \({\pm }\) 0.0078

0.1750 \({\pm }\) 0.0084

0.1850 \({\pm }\) 0.0078

0.1808 \({\pm }\) 0.0029

ECL2

2.2200

0.2230\(^{**-}\) \({\pm }\) 0.0064

0.1840 \({\pm }\) 0.0091

0.1930 \({\pm }\) 0.0084

0.1920 \({\pm }\) 0.0016

ECL3

2.4500

0.2400\(^{**-}\) \({\pm }\) 0.0033

0.1930 \({\pm }\) 0.0089

0.1990 \({\pm }\) 0.0064

0.1971 \({\pm }\) 0.0022

APO

3p0g

ECL1\(^{**+}\)

1.4983

0.1935\(^{**-}\) \({\pm }\) 0.0005

0.1702 \({\pm }\) 0.0049

0.1718 \({\pm }\) 0.0021

0.1641 \({\pm }\) 0.0014

ECL2

1.5356

0.1996\(^{*-}\) \({\pm }\) 0.0056

0.1821 \({\pm }\) 0.0055

0.1828 \({\pm }\) 0.0047

0.1767 \({\pm }\) 0.0012

ECL3

1.3340

0.2003\(^{**-}\) \({\pm }\) 0.0071

0.1929 \({\pm }\) 0.0060

0.1949 \({\pm }\) 0.0046

0.1895\({\pm }\) 0.0015

FA

2rh1

ECL1\(^{*+}\)

1.4525

0.1962\(^{**-}\) \({\pm }\) 0.0028

0.1681 \({\pm }\) 0.0060

0.1679 \({\pm }\) 0.0054

0.1692 \({\pm }\) 0.0054

ECL2

1.8524

0.2064\(^{**-}\) \({\pm }\) 0.0065

0.1730 \({\pm }\) 0.0059

0.1720 \({\pm }\) 0.0058

0.1728 \({\pm }\) 0.0053

ECL3

1.4563

0.2133\(^{**-}\) \({\pm }\) 0.0033

0.1823 \({\pm }\) 0.0041

0.1819 \({\pm }\) 0.0041

0.1816 \({\pm }\) 0.0040

FA

3p0g

ECL1\(^{**+}\)

1.4513

0.2119\(^{**-}\) \({\pm }\) 0.0041

0.1679 \({\pm }\) 0.0061

0.1682 \({\pm }\) 0.0033

0.1618 \({\pm }\) 0.0031

ECL2

1.6783

0.2340\(^{**-}\) \({\pm }\) 0.0012

0.1776 \({\pm }\) 0.0070

0.1765 \({\pm }\) 0.0051

0.1715 \({\pm }\) 0.0031

ECL3

1.7785

0.2157\(^{**-}\) \({\pm }\) 0.0022

0.1860 \({\pm }\) 0.0066

0.1845 \({\pm }\) 0.0059

0.1806 \({\pm }\) 0.0023

PIA

2rh1

ECL1\(^{*+}\)

1.9613

0.1934\(^{**-}\) \({\pm }\) 0.0026

0.1711 \({\pm }\) 0.0039

0.1690 \({\pm }\) 0.0027

0.1664 \({\pm }\) 0.0066

ECL2

2.3271

0.2017\(^{**-}\) \({\pm }\) 0.0019

0.1732 \({\pm }\) 0.0033

0.1727 \({\pm }\) 0.0033

0.1680 \({\pm }\) 0.0063

ECL3

2.5942

0.2235\(^{**-}\) \({\pm }\) 0.0022

0.1918 \({\pm }\) 0.0040

0.1893 \({\pm }\) 0.0035

0.1831 \({\pm }\) 0.0055

PIA

3p0g

ECL1\(^{*+}\)

2.1627

0.1945\(^{**-}\) \({\pm }\) 0.0021

0.1647 \({\pm }\) 0.1793

0.1577 \({\pm }\) 0.1771

0.1604 \({\pm }\) 0.1563

ECL2

2.0666

0.2137\(^{**-}\) \({\pm }\) 0.0022

0.1690 \({\pm }\) 0.2008

0.1598 \({\pm }\) 0.1974

0.1633 \({\pm }\) 0.1773

ECL3

2.8734

0.2234\(^{**-}\) \({\pm }\) 0.0025

0.1924 \({\pm }\) 0.2146

0.1843 \({\pm }\) 0.2212

0.1878 \({\pm }\) 0.2042

Mean

ECL1\(^{*+}\)

1.7677

0.2056\(^{**-}\) \({\pm }\) 0.0033

0.1695 \({\pm }\) 0.0348

0.1699 \({\pm }\) 0.0331

0.1671 \({\pm }\) 0.0293

ECL2

1.9467

0.2131\(^{**-}\) \({\pm }\) 0.0040

0.1765 \({\pm }\) 0.0386

0.1761 \({\pm }\) 0.0374

0.1741 \({\pm }\) 0.0325

ECL3

2.0811

0.2194\(^{**-}\) \({\pm }\) 0.0034

0.1897 \({\pm }\) 0.0407

0.1890 \({\pm }\) 0.0410

0.1866 \({\pm }\) 0.0366

  1. Statistics of significantly worse\(^{(-)}\) and significantly better\(^{(+)}\) differences with p \(< 0.05^{*}\) and p\(< 0.01^{**}\) have been included. The symbols in the first column indicate significantly better differences between ECL regions, while, in the model results columns, they indicate significantly worse differences between models. The Mean row at the bottom indicates mean errors across regions. Bold values show the minimum accross regions and models.
Back to article page