Extended Data Table 2 |. Ranking of fitness models disregarding subclonal composition of tumors.

Fitness models from Extended Data Table 1 evaluated without accounting for subclonal composition of tumors on the same three cohorts, Van Allen et al. (n=103), Snyder et al. (n=64) and Rizvi et al. (n=34). As in Extended Data Table 1, we report parameters used for predictions and error bars for these parameters (Methods). Parameter 𝜏 is not a free parameter when disregarding subclonal composition of tumors (Methods) and is not reported. We report log-rank test scores for all models and the log-rank test p-value for models with significant patient segregation (p<0.05). The significant models are highlighted: yellow for models significant in a single cohort, orange for models significant across two cohorts, and red for models significant in all three cohorts.

Van Allen et al., Melanoma anti-CTLA-4	Parameters trained on Snyder et al. dataset				Log-rank test
Models	a		k		Score		Significance		Equation
	Mean	Std	Mean	Std	Mean	Std
Ax R	18.5	±0.152	0.59845	±0.001	0.61	±0.24			(2)
Partial models:
A	-	-	-	-	0.05	±0.03			(12)
R	18.2	±2.995	4.54981	±0.001	0.2	±0.03			(13)
1/kdMxR	26.6	±1.609	1.34421	±0.001	1.41	±0.61			(14)
1/kdM	-	-	-	-	0.69	±0.23			(14)
kdWxR	23.7	±2.593	3.83397	±0.001	1.35	±0.34			(15)
kdW	_	_	_	_	0.46	±0.2			(15)
Alternative models:
Neoantigen load	_	_	_	_	0.24	±0.12			(17)
A x R, sum over neoantigens	25.1	±3.308	4.89176	±0.001	1.57	±0.39			(16)
Ax R, alignments at positions 3–8	21.9	±2.871	3.1276	±0.001	0.35	±0.36			(2)
Ax R, negative IED3 assays	14.4	±2.58	2.20223	±0.001	0.04	±0.03			(2)
A x R, all neoantigens	29.2	±2.892	3.88642	±0.001	0.24	±0.19			(2)
Snyder et al., Melanoma anti-CTLA-4	Parameters trained on Van Allen et al. dataset				Log-rank test
Models	a		k		Score		Significance		Equation
	Mean	Std	Mean	Std	Mean	Std
Ax R	26.4	±0.892	1.0851	±0.001	6.55	±0.9	0.01047	*	(2)
Partial models:
A	-	-	-	-	4.44	±0.68	0.03507	*	(12)
R	29.7	±4.829	2.51962	±0.001	1.26	±0.34			(13)
1/kdM x R	26	±1.929	0.82074	±0.001	3.65	±0.81			(14)
1/kdM	_	_	_	_	3.44	±0.61			(14)
kdWxR	26.9	±2.735	3.73387	±0.001	1.67	±0.22			(15)
kdW	-	-	-	-	3.11	±0.75			(15)
Alternative models:
Neoantigen load	-	-	-	-	0.42	±0.21			(17)
A x R, sum over neoantigens	27	±3.005	5.08369	±0.001	1.63	±0.53			(16)
Ax R, alignments at positions 3–8	30	±4.023	1.35553	±0.001	2.73	±0.65			(2)
Ax R, negative IED3 assays	36	±9.57	10.1531	±0.001	0.23	±0.65			(2)
Ax R, all neoantigens	26	±1.95	5.22216	±0.001	0.59	±0.92			(2)
Rizvi et al., Lung anti-PD-1	Parameters trained on Van Allen et al. and Snyder et al. datasets				Log-rank test
Models	a		k		Score		Significance		Equation
	Mean	Std	Mean	Std	Mean	Std
Ax R	27	±0.787	1.00032	±0.001	6.48	±1.14	0.01088	*	(2)
Partial models:
A	-	-	-	-	4.65	±1.17	0.03099	*	(12)
R	19.6	±3.355	4.29127	±0.001	1.53	±0.29			(13)
1/kdM x R	21	±2.027	0.53498	±0.001	0.02	±0.07			(14)
1/kdM	-	-	-	-	0.17	±0.13			(14)
kdWxR	23	±2.737	5.37707	±0.001	10.48	±1.71	0.00121	***	(15)
kdW	-	-	-	-	4.49	±0.75	0.03416	*	(15)
Alternative models:
Neoantigen load	-	-	-	-	4.93	±1.15	0.02639	*	(17)
A x R, sum over neoantigens	25	±4.316	4.03113	±0.001	3.09	±1.09			(16)
Ax R, alignments at positions 3–8	22	±3.042	5.2228	±0.001	2.3	±0.82			(2)
Ax R, negative IED3 assays	14.5	±2.806	1.90577	±0.001	1	±0.45			(2)
Ax R, all neoantigens	29	±4.469	1.87092	±0.001	0.75	±0.62			(2)