Table 1 Convergence diagnostics comparing three chains of 5000 samples run for all reactions in the Poolman, Arnold, and Dal’Molin models using the ACHR, OPTGP, and CHRR algorithm with the indicated thinning

From: Flux sampling is a powerful tool to study metabolism under changing environmental conditions

Model

Sampler

Thinning

Convergence diagnostics

Nmax (Raftery & Lewis)

% I > 5 (Raftery & Lewis)

% Z > 1.28 (Gweke)

% IPSRF < 0.9 or IPSRF > 1.1

Poolman

ACHR

100

2862482

53

28

30

1000

1557990

46

23

16

10,000

1083360

41

20

14

OPTGP

100

1664187

53

22

13

1000

1099330

45

21

11

10,000

1717480

40

17

6

CHRR

100

303880

15

14

0

1000

27906

0

12

0

10,000

4636

0

10

0

Arnold

ACHR

100

385177

49

39

9

1000

312384

12

34

3

10,000

323570

9

23

0

OPTGP

100

910210

45

36

7

1000

1538190

12

30

3

10,000

1679346

9

32

17

CHRR

100

122760

17

24

0

1000

17508

0

21

0

10,000

4198

0

19

0

Dal’Molin

ACHR

100

41951912

37

13

17

1000

4982459

28

12

10

10,000

1544792

23

10

6

OPTGP

100

1834104

37

17

14

1000

1891665

27

11

6

10,000

1700907

22

11

3

CHRR

100

550776

10

16

3

1000

387456

4

16

3

10,000

226856

3

13

2

  1. Three samples sets (chains) were computed for each reaction using the indicated models, algorithms and thinnings. Nmax indicates the maximum number of total samples suggested by the Raftery & Lewis diagnostic. % I > 5 and % z > 1.28 indicate the average percentage of reactions which do not pass the Raftery & Lewis and Gweke convergence diagnostics according to the three chains. The three chains were used for each reaction in order to compare within and between sample differences as indicated by the interval-based potential scale reducing factor (IPSRF) diagnostic. The percentage of reactions with IPSRF < 0.9, IPSRF > 1.1 is shown