Development of CYP3A4 - in silico Metabolic Stability 2011.07.18 Jeong Seung Hwa

GFA-MLR Km – QSAR model Aliphatic Hydroxylation r2 = 0.94 q2=0.90 Aromatic Hydroxylation r2 = 0.93 q2=0.86 N-dealkylation r2 = 0.83 q2=0.87 O-dealkylation r2 = 0.97 q2=0.88

GFA-MLR Vmax – QSAR model Aliphatic Hydroxylation r2 = 0.86 q2=0.82 Aromatic Hydroxylation r2 = 0.91 q2=0.86 N-dealkylation r2 = 0.92 q2=0.83 O-dealkylation r2 = 0.90 q2=0.85

GFA-MLR Metabolic stability prediction Ordinary Differential Equation (ODE) 오일러 Euler Method

Simulation in silico Metabolic Stability (4개 - only 3A4) r2 = 0.982 J. Pharmacol.Exp.Ther. 283 (1997) 46–58

Simulation in silico Metabolic Stability with Single Reaction (17개: 3A4) r2 = 0.6344 promazine Buspirone r2 = 0.8267 J. Pharmacol.Exp.Ther. 283 (1997) 46–58

Try for different method QSAR [GFA-MLR] Classification – 문제점 : stable/ unstable 기준 모호 QSAR [ non-linear] A Back propagation Neural Network [GNN] 175 compound reaction Vmax Km aliphatic hydroxylation 64 50 aromatic hydroxylation 29 14 N-dealkylation 80 57 O-dealkylation 22

Km GNN Descriptor [ by GFA] aliphatic hydroxylation Ea Graph_Petitjean No_H_bond_donors SKlogD_value aromatic hydroxylation Total_structure_connectivity_index Valence_bound_charge_index_03 N-dealkylation Fraction_of_2D_VSA_hydrophobic_sat VChi_05_path E_state_SaasC O-dealkylation E_state_SaaN AlogP98_value Initial equation length Maximum equation length

Vmax GNN Ea Descriptor [ by GFA] aliphatic hydroxylation AlogP98_value Eccentric_adjacency_index AI_SsssCH aromatic hydroxylation Total_structure_connectivity_index Fraction_of_2D_VSA_Hbond_acceptor N-dealkylation SKlogD_value VChi_05_path AI_SssNH O-dealkylation Fraction_of_2D_VSA_hydrophobic

GNN A Back propagation Neural Network (BNN) Aliphatic hydroxylation [Km] Ea Graph_Petitjean No_H_bond_donors SKlogD_value 4-4-1 r2 = 0.801

GNN Description Graph_Petitjean SKlogD = logP = Σniai (Graph diameter – graph radius ) / graph diameter Graph diameter : 거리행렬 (distance matrix)의 요소값 중 가장 큰 값 Graph radius : 거리행렬 (distance matrix)의 행에서 가장 큰 값들 중 그 값이 가장 작은 행의 값 [ Petitjean,M. J. Chem. Inf. Comput. Sci. 1992, 32, 331. ] SKlogD = logP = Σniai ni 는 분자내에 해당 atom type i 의 수를 나타내며, ai 는 해당 atom type의 기여도. logP는 섞이지 않는 octanol과 water층에 해당 화합물을 투여하여 각각 분배되는 분배계수( logP값이 클수록 지 질층을 대표하는 octanol층에 화합물이 많이 분배되는 것으로 소수성이 큼. 반면에 logP값이 작을수록 물층에 많 이 분배되므로, 친수성이 큼을 나타냄)

GNN A Back propagation Neural Network (BNN) Aromatic hydroxylation [Km] Ea Total_structure_connectivity_index Valence_bound_charge_index_03 3-3-1 r2 = 0.941

GNN Description charge index (Gk) bound charge index (Jk) k : 각 path의 길이 , dij : distance matrix 의 요소, δ(k, dij) : Kronecker delta dij =k인 경우는 1, 그렇지 않은 경우는 0을 표시 그러므로 Gk 는 dij=k인 모든 CTij항들의 합. bound charge index (Jk) 각 결합에 대한 전하 이동의 평균값을  valence charge index (Gkˇ , Jkˇ ) Charge index 와 유사하지만 인접행렬 A 대신에 A ˇ 를 사용. 행렬 A ˇ 는 일반적으로 인접행렬 A와 동일, 행렬의 대 각선에 있는 요소값은 A행렬의 경우 모두 0이지만, A ˇ 는 탄소와의 Pauling 전기음성도 차이 (electronegativity (EN) difference)로 얻어진 값. [ Galvez, et al. J. Chem. Inf. Comput. Sci. 1994, 34, 520. ]

GNN A Back propagation Neural Network (BNN) N-dealkylation [Km] 4-4-1 Fraction_of_2D_VSA_hydrophobic_sat VChi_05_path E_state_SaasC 4-4-1 r2 = 0.8365

GNN Description Electrotopological state indices (E-state) 분자내에서 원자-원자의 정전기적 상호작용과 원자 들의 위상적 환경 (연결)에 대한 정보 제공. intrinsic state (Ii) δ i(vertex degree) : 원자에 결합되어 있는 edge의 수, 인접행렬에서 i번째 행의 모든 요소의 합 Intrinsic state : 일반적으로 원자의 원자가상태 (valence state)의 전기음성도 (electro-negativity)에 비례하며, 단 순하게는 해당 원자에 대하여 σ결합 수에 대한 π전자 및 비공유 전자 쌍의 비율. Electrotopological state index (E-state index: Si ) rij = 원자간의 거리, 그래프 거리에 1을 더한 값 (rij = dij + 1). 즉 해당 원자의 intrinsic state (Ii)값에 이 해당 원자에 대하여 분자내의 모든 원자들의 영향을 반영한 perturbation term ( Δ Ii)을 더해서 계산. Atom-type E-state indices: 원자의 혼성 (hybridization) 또는 원 자들 사이의 연결정보를 고려한 E-state [ Hall,L.H. and Kier,L.B. J. Chem. Inf. Comput. Sci. 1995, 35, 1039. (2) Hall,L.H. and Kier,L.B. J. Chem. Inf. Comput. Sci. 1995, 35, 1074. ]

GNN A Back propagation Neural Network (BNN) O-dealkylation [Km] 3-3-1 E_state_SaaN AlogP98_value 3-3-1 r2 = 0.9986

GNN A Back propagation Neural Network (BNN) Aliphatic hydroxylation [Vmax] Ea AlogP98_value Eccentric_adjacency_index AI_SsssCH 4-4-1 r2 = 0.7514

GNN Description Eccentric_adjacency_index 각 원자의 인접한 원자들의 정점도 합 과 그 원자의 eccentricity와의 비율을 각 원자들에 대하여 총합계한 값. 정점도 (vertex degree) : 원자에 결합되어 있는 edge의 수 , 인접행렬에서 i번째 행의 모든 요소의 합. Oi : 각 원자 (정점 vi)에 인접한 원자들의 정점도 (vertex degree)의 합 E(i) : 정점 vi의 eccentricity (화학그래프에서 정점 vi로부터 가장 먼 정점까지의 거리). [ Gupta,E. et al. J. Comp. Aided Mol. Des. 2001, 15, 671. ] Fraction of 2D-VSA hydrophobic 전체 VDW 표면적에 대한 소수성 원자들의 VDW 표면적 비율 Edge 모서리

GNN Description [Atom-type AI topological indices (AI)] 여기서 δ i : i 번째 원자의 정점도 (vertex degree), σi : 거리행렬(distatnce matrix)에서의 거리도 (distancedegree: 거리행렬에서 i 번째 행의 총합). Фi (j) : i 번째 원자에 대하여 다른 원자들이 영향을 주는 perturbing term [ (1) Ren,B. Comp. Chem. 2002, 26, 223. (2) Ren.B. J. Chem. Inf. Comput. Sci. 2002, 42, 858. ]

GNN A Back propagation Neural Network (BNN) Aromatic hydroxylation [Vmax] Ea AlogP98_value Total_structure_connectivity_index Fraction_of_2D_VSA_Hbond_acceptor 4-4-1 r2 = 0.9646

GNN A Back propagation Neural Network (BNN) N-dealkylation [Vmax] SKlogD_value VChi_05_path AI_SssNH 4-4-1 r2 = 0.7025

GNN A Back propagation Neural Network (BNN) O-dealkylation [Vmax] Eccentric_adjacency_index Fraction_of_2D_VSA_hydrophobic 3-3-1 r2 = 0.8655

Current & Further Study CSB Further study Current & Further Study Model optimization 논문작성

Aliphatic Hydroxylation/ N-deaklyation / O-dealkylation CSB Aliphatic Hydroxylation/ N-deaklyation / O-dealkylation [Aliphatic Hydroxylation] Hydrogen abstraction Oxygen rebound [N-deaklyation / O-dealkylation ] Mechanism적으로 접근 The ionization potential of the radical may be a quantitative measure of this stabilization, since the energy of the singly occupied orbital is lowered through resonance. e - abstraction Hydrogen abstraction Oxygen rebound

Aromatic Hydroxylation CSB Aromatic Hydroxylation ΔHact = 21.91 + 0.61(ΔHreac) Aliphatic hydroxylation : an initial hydrogen abstraction followed by the so called oxygen rebound Aromatic hydroxylation : probably initiated by an electrophilic attack by the FeO3+ complex . The reaction then proceeds either through a 1,2-migration or via an epoxidation reaction followed by a 1,2-migration an initial electrophilic attack on the ð-system of the aromatic ring to produce a tetrahedral radical or cationic ó-complex. due to the resonance stabilization of the radical cation in the aromatic system. If the oxidation of primary amines proceeds via an initial abstraction of an electron from the nitrogen atom, the rate would be expected to show a dependence on the Hammett electronic parameter, u(rou), provided the oxidation occurs prior to or during the rate-determing step the rate of hydroxylation of the aromatic system was about 10 times that of the aliphatic system. the hydrogen atom abstractions are predicted by AM1 to have barriers around 20 kcal/mol, whereas aromatic oxidation has barriers around 10 kcal/mol. The linear correlation of aromatic activation energies and the heats of reaction for formation of a tetrahedral intermediate by methoxy radical. DMD 30:7–12, 2002

CSB Introduction Metabolic stability prediction Km, Vmax prediction by QSAR Model Ex) CYP 450 3A4 E + S <----> ES ----> E + P Vmax = k2 [E]0 (k2 ≈ Activation Energy, Vmax ≈ C1(Activation Energy) + C2) KM= k-1 + k2 = Kd + k2 k1 k1 if k2 is the rate-limitings step (k2 >> k1) KM ≈ Kd + k2 ≈ Kd + Vmax KM ≈ C1(Binding Affinity) + C2 (Activation Energy) + C3 k1 k2 K-1 Michaelis-Menten Kinetics Vmax = the maximum metabolism rate capable by enzymes Km = the Michaelis-Menten Constant K1,k-1 : substrate binding , k2 = catalytic step(kcat) Km = The rate of ES formation = the rate of ES break down

CSB Introduction Metabolic stability prediction Km, Vmax prediction by QSAR Model Ex) CYP 450 3A4 E + S <----> ES ----> E + P Keq = k1 = e -ΔΔG/RT k-1 kp= k2 = Ae -Ea/RT KM = k-1 + k2 = 1/Keq + kp k1 KM = e ΔΔG/RT + Ae -Ea/RT k1 k2 K-1 From Boltzmann distribution law - binding affinity Arrhenius equation - rate-determining step Binding affinity Activation energy J. Chem. Inf. Model., 2008, 48 (5), pp 1074–1080