Table 6.

Determining the reduction path for the graph-based reduction approach.

Step	Context 1	Context 2	Log likelihood difference	Performed	Model
1	CXC	TXC	872.10	YES	GTR15C
2	GXG	TXG	1064.87	YES	GTR14C
3	GXT	TXC	1300.93	YES	GTR13C
4	CXC	GXC	1524.96	YES	GTR12C
5	AXC	GXC	1750.50	NO
6	GXC	TXC	1777.24	SKIP
7	AXT	CXG	1868.77	YES	GTR11C
8	CXC	GXT	1925.79	SKIP
9	CXA	GXT	1928.98	YES	GTR10C
10	GXC	GXT	1940.97	SKIP
11	GXA	TXG	1961.01	YES	GTR9C
12	AXC	GXA	1964.73	YES	GTR8C
13	GXA	GXG	1973.80	SKIP
14	AXC	CXG	1983.92	YES	GTR7C
15	AXC	TXG	2043.25	SKIP
16	CXA	GXC	2047.72	SKIP
17	AXC	GXG	2211.16	SKIP
18	CXG	TXG	2302.08	SKIP
19	CXG	GXG	2341.46	SKIP
20	CXG	GXC	2343.32	YES	GTR6C
21	AXC	AXT	2352.48	SKIP
22	GXT	TXA	2363.72	YES	GTR5C
...	...	...	...	...	...

The graph-based reduction approach constructs the optimal model (GTR8C in Table 8) for the Ancestral Repeats dataset in 12 iterations (first column). The second and third column show which 2 contexts (or clusters) are proposed for merging; the fourth column shows the difference in log likelihood between the full 16-contexts model and the resulting 15-contexts model should only those 2 contexts given in the second and third column be merged; the fifth column shows the decision on the proposed merge (YES: the merge is performed; NO: the merge is not performed due to a lower cost alternative; SKIP: the merge is already present in the current clustering); the sixth column shows the resulting model when a merge operation is performed.