Table C1.

Comparison of model parameters for phenotypes and Hapgen-based simulations using the model parameters (with ${\sigma }_0=1$) to produce the underlying distribution of effects

Phenotype	π 1	${\mathbf{\sigma }}_b^{\mathbf{2}}$	${\mathbf{\sigma }}_c^2$	S	${\mathbf{\pi }}_1{p}_{c1}$	mc	wc	${\mathbf{\sigma }}_d^2$	${\mathbf{\pi }}_1{p}_{d1}$	md	wd	${\mathbf{\sigma }}_0^2$
SCZ 2014	5.28e−2	1.4e−6	3.6e−5	−0.52	3.7e−3	87	352	1.4e−4	3.7e−4	519	7	1.07
Simulation	1.05e−2	2.9e−6	3.2e−5	−0.42	2.6e−3	59	174	1.1e−4	6.8e−3	559	119	1.08
BIP	5.91e−2	1.2e−6	4.5e−5	−0.40	4.0e−3	102	414	—–	—–	—–	—–	1.01
Simulation	6.86e−3	1.6e−5	3.9e−5	−0.40	5.1e−3	64	162	—–	—–	—–	—–	1.07
CD	9.55e−4	5.0e−5	5.5e−4	−0.64	1.9e−4	176	604	7.6e−2	2.0e−5	—–	—–	1.14
Simulation	6.29e−4	1.1e−4	4.4e−4	−0.84	1.8e−4	127	409	2.6e−2	9.8e−3	20	50	1.05
UC	1.16e−3	3.6e−5	4.0e−4	−0.67	1.9e−4	173	627	8.0e−2	1.6e−5	—–	—–	1.12
Simulation	9.50e−4	5.5e−5	4.0e−4	−0.65	1.8e−4	67	214	6.0e−4	9.5e−3	—–	—–	1.04
CAD	1.88e−3	1.1e−5	9.2e−5	−0.51	1.3e−4	171	683	5.3e−3	2.5e−5	102	7	0.92
Simulation	4.01e−3	5.0e−6	3.2e−3	−0.33	6.4e−6	45	143	1.8e−4	2.4e−5	—–	—–	1.02
AD Chr19	4.34e−4	1.0e−4	6.1e−3	−0.57	3.2e−4	35	89	—–	—–	—–	—–	1.09
Simulation	3.88e−4	7.4e−4	1.8e−3	−0.60	3.9e−4	—–	—–	—–	—–	—–	—–	1.11
AD NoC19	1.05e−3	1.8e−5	2.6e−4	−0.52	3.0e−5	264	6	—–	—–	—–	—–	1.04
Simulation	2.52e−3	8.8e−6	9.8e−5	−0.78	5.3e−5	708	757	—–	—–	—–	—–	1.00
ALS Chr9	1.12e−2	7.3e−6	3.9e−3	−0.01	2.0e−5	106	6	—–	—–	—–	—–	0.99
Simulation	5.23e−3	2.0e−5	1.2e−2	1.00	8.4e−6	144	7	—–	—–	—–	—–	1.03
Edu	1.43e−2	1.7e−6	7.8e−6	−0.44	6.4e−3	111	339	8.5e−5	2.8e−3	441	7	0.94
Simulation	7.86e−3	4.4e−6	1.4e−5	−0.45	3.4e−3	62	186	1.0e−4	4.7e−3	1031	8	1.03
IQ 2018	1.27e−2	7.5e−7	6.2e−6	−0.51	4.8e−3	122	309	3.6e−5	3.0e−2	561	6	1.17
Simulation	4.52e−3	8.4e−6	1.6e−5	−0.49	2.1e−3	59	112	2.6e−5	9.2e−2	421	15	1.02
Height 2010	1.02e−3	4.1e−5	2.0e−4	−0.44	2.0e−4	322	1243	—–	—–	—–	—–	0.90
Simulation	1.02e−3	5.2e−5	2.2e−4	−0.57	1.5e−4	152	478	—–	—–	—–	—–	1.02
Height 2014	1.15e−3	3.7e−5	1.6e−4	−0.46	2.4e−4	242	929	—–	—–	—–	—–	1.57
Simulation	8.26e−4	6.7e−5	3.9e−4	−0.40	3.5e−5	1363	193	—–	—–	—–	—–	1.05
HDL	2.54e−3	1.1e−5	4.5e−4	−0.79	3.6e−5	143	599	2.2e−2	1.5e−5	66	7	0.91
Simulation	9.39e−4	3.1e−5	1.1e−3	−0.70	2.3e−5	68	217	9.8e−4	1.6e−4	—–	—–	1.02
LDL	5.84e−3	3.3e−6	2.4e−4	−0.52	5.1e−5	336	1417	7.3e−3	1.9e−6	346	6	0.92
Simulation	3.86e−3	5.5e−6	3.3e−4	−0.58	3.1e−5	2046	273	—–	—–	—–	—–	1.03
BMI GIANT 2015	1.54e−3	2.2e−5	4.5e−4	0.00	6.8e−6	288	12	—–	—–	—–	—–	0.85
Simulation	1.09e−3	2.4e−5	7.9e−4	0.73	4.3e−6	440	7	—–	—–	—–	—–	1.02
TC	1.15e−3	1.7e−5	6.2e−	−0.97	2.4e−5	140	583	2.9e−4	7.1e−4	—–	—–	0.92
Simulation	7.16e−4	2.9e−5	2.1e−4	−0.95	4.8e−5	255	822	5.0e−3	6.7e−4	83	108	1.01

π ₁ is the overall proportion of the 11 million SNPs from the reference panel that are estimated to be causal. ${\pi }_1\times p_c(L⩾1)$ is the total prior probability multiplying the “c” Gaussian, which has variance ${\sigma }_c^2{H}^S$, where H is the reference SNP heterozygosity. Note that $p_c(L)$ is just a sigmoidal curve, and can be characterized quite generally by three parameters: the value $p_{c1}\equiv p_c(1)$ at L = 1; the total LD value L = m_c at the mid-point of the transition, i.e. $p_c(m_c)=p_{c1}/2$ (see the middle gray dashed lines in Figure 1, which shows examples of π₁ times the function $p_c(L)$); and the width w_c of the transition, defined as the distance (in L) between where the curve falls to 95% and 5% of $p_{c1}$ (distance between the flanking red dashed lines in Figure 1). The “d” Gaussian is similarly defined (but without the H^S dependence in the variance). Note that for AD Chr19, AD NoC19, and ALS Chr9, π₁ is the fraction of reference SNPs on chromosome 19, on the autosome excluding chromosome 19, and on chromosome 9, respectively. See Figures 6 and 7, and Supplementary Figures S1 and S2. Ninety-five percent confidence intervals for the parameter estimates for the real phenotypes are in Appendix Tables E1–E3. The simulations were run in a separate processing stream; we did not additionally calculate confidence intervals for the simulation parameter estimates, which would have required substantial code modification and extensive processing. However, we expect on heuristic grounds that they would be similar to those estimated for the real phenotypes.