The volume of data worldwide doubles every three years. Application of machine learning (ML) and artificial intelligence (AI) techniques to ever-growing biochemical databases allows for building predictive models for candidate drugs’ properties, including activity, toxicity, and synergies.
In drug design pipelines, such insights help formulate initial hypotheses and narrow down the number of options to consider, significantly limiting the costs and time of experimental campaigns.
In patient-oriented applications, personalized assessment of efficacy and toxicity for individual drugs as well as their combinations help optimize therapeutic outcomes.
Our flagship AI services are described below.
Protein targets
Our AI-driven model for protein target prediction is designed to support the fast-paced environment of drug discovery. Our model streamlines the prediction of compound activities across a broad spectrum of protein targets, enhancing the efficiency of virtual assays and screening processes.
By facilitating the rapid evaluation of millions of compounds, it offers a significant advantage in both time and resources. The model provides small molecule activity estimates with performance exceeding state of the art competition.
Access is available through ai.inventro.co and personalized tokens are required for entry. To explore the research capabilities of a model with a free trial, please reach out to us.
Protein target prediction
Protein target prediction is a critical component of modern drug design and pharmaceutical research. Computational methods provide fast and cost efficient solutions for obtaining initial hypotheses in areas such as virtual screening, drug repurposing or side effects prediction. Their predictive value constantly increases, benefitting from exponential growth of available data and computer power. Currently, given naturally imperfect reproducibility of experimental activity assessments, virtual assays are estimated to provide overall comparable performance to their in-vitro counterparts, while offering significantly broader insights with only a fraction of cost and effort.
Our activity predictions are based on a neural network model trained on small molecule activity data available in the public domain, particularly these deposited in the the ChEMBL database. The model is regularly updated to include the newest records. After quality-based filtering of available data and selection of protein targets with favourable prediction accuracy, the model provides small molecule activity estimates against over a thousand individual protein targets across several species, with performance exceeding state of the art competition. High computational efficiency allows screening millions of compounds per 24 hours.
The model employs input queries in the form of commonly used molecular representations (e.g., SMILES, MOL2, SDF) of compound database. Each individual molecular structure undergoes standardisation, transformation into a vectorised representation, and subsequent submission to a neural network-based activity predictor.
For each compound, the primary output includes an estimate of affinity against all targets together with the assessment of response selectivity and sensitivity.
Virtual assays
Responses from individual targets are grouped according to target class into virtual assays panels that mirror those widely used in in-vitro screening assays. For each compound they provide instant information concerning possible area of application, off-target prediction, as well as potential safety concerns.
Protein kinases
Kinase binding assays play a crucial role in drug design, particularly in the development of targeted therapies. Our virtual kinase binding assay provides for rapid selection of putative protein kinase binders, as well as for initial selectivity profiling. It comprises 318 unique targets (with 302 human proteins), including 227 serine/threonine kinases and 74 tyrosine kinases.
| Assay type | Protein genes |
| Serine/threonine kinases | AAK1, ACVR1, ACVR1B, ACVRL1, AKT1, AKT2, AKT3, ARAF, ATR, AURKA, AURKB, AURKC, BCR, BMP2K, BMPR1A, BRAF, BRSK1, CAMK1, CAMK1D, CAMK2B, CAMK2D, CAMK2G, CAMKK2, CDC42BPA, CDC7, CDK1, CDK12, CDK13, CDK19, CDK2, CDK4, CDK5, CDK6, CDK7, CDK8, CDK9, CDPK1, CHEK1, CHEK2, CHUK, CSNK1A1, CSNK1D, CSNK1E, CSNK1G1, CSNK1G2, CSNK1G3, CSNK2A1, CSNK2A2, DAPK3, DCLK1, DYRK1A, EEF2K, EIF2AK1, EIF2AK3, EIF2AK4, GAK, GRK2, GRK5, GSK3-BETA, GSK3A, GSK3B, HASPIN, HIPK2, HIPK4, IKBKB, IKBKE, ILK, IRAK1, IRAK4, LATS1, LIMK1, LIMK2, LRRK2, MAP3K11, MAP3K12, MAP3K14, MAP3K20, MAP3K5, MAP3K7, MAP3K8, MAP3K9, MAP4K1, MAP4K2, MAP4K3, MAP4K4, MAP4K5, MAPK1, MAPK10, MAPK11, MAPK12, MAPK13, MAPK14, MAPK3, MAPK7, MAPK8, MAPK9, MAPKAPK2, MAPKAPK5, MARK2, MARK3, MELK, MINK1, MKNK1, MKNK2, MTOR, NEK2, NEK4, NUAK1, PAK1, PAK4, PDK1, PDK2, PDK3, PDK4, PDPK1, PHKG2, PIM1, PIM2, PIM3, PKMYT1, PKN1, PKN2, PKNB, PLK2, PLK3, PLK4, PRKAA1, PRKAA2, PRKACA, PRKACB, PRKACG, PRKCA, PRKCB, PRKCD, PRKCE, PRKCG, PRKCH, PRKCI, PRKCQ, PRKCZ, PRKD1, PRKD2, PRKD3, PRKG1, PRKG2, PRKX, RAF1, RIPK1, RIPK2, RIPK3, ROCK1, ROCK2, RPS6KA1, RPS6KA2, RPS6KA3, RPS6KA5, RPS6KA6, RPS6KB1, SGK1, SGK2, SIK2, SLK, SRPK1, STK10, STK16, STK17A, STK17B, STK3, STK4, TAOK1, TAOK3, TBK1, TGFBR1, TGFBR2, TNIK, TNNI3K, TSSK1B, ULK1 |
| Tyrosine kinases | ABL1, ABL2, ALK, AXL, BLK, BMX, BTK, CSK, DDR1, DDR2, EGFR, EPHA2, EPHA4, EPHB4, ERBB2, ERBB4, FER, FES, FGFR1, FGFR2, FGFR3, FGFR4, FGR, FLT1, FLT3, FLT4, FRK, FYN, HCK, IGF1R, INSR, ITK, JAK1, JAK2, JAK3, KDR, KIT, LCK, LTK, LYN, MERTK, MET, MST1R, NTRK1, NTRK2, NTRK3, PDGFRA, PDGFRB, PTK2, PTK2B, PTK6, RET, ROS1, SRC, SRMS, SYK, TEC, TEK, TNK1, TNK2, TXK, TYK2, TYRO3, WEE1, YES1, ZAP70 |
GPCRs
G-Protein Coupled Receptors (GPCRs) are a key group of protein targets owing to their abundance, diversity, clinical relevance, and well-established druggability. At the some time, GPCRs pose unique challenges for in vitro screening approaches due to their integral membrane protein nature. Our virtual GPCR binding assay provides quick and easily accessible initial assessment of small molecule affinity and selectivity across a broad spectrum of receptors. The panel includes 230 unique GPCR targets (with 154 human proteins).
| Assay type | Protein genes |
| GPCRs | ACKR3, ADORA1, ADORA2A, ADORA2B, ADORA3, ADRA1A, ADRA1B, ADRA1D, ADRA2A, ADRA2B, ADRA2C, ADRB1, ADRB2, ADRB3, AGTR1, AGTR1B, AGTR2, APLNR, AVPR1A, AVPR1B, AVPR2, BDKRB1, BDKRB2, C3AR1, C5AR1, CALCR, CALCRL, CCKAR, CCKBR, CCR1, CCR2, CCR3, CCR4, CCR5, CHRM1, CHRM2, CHRM3, CHRM4, CHRM5, CNR1, CNR2, CXCR1, CXCR2, CXCR3, CXCR4, CYSLTR1, CYSLTR2, DRD1, DRD2, DRD3, DRD4, DRD5, EDNRA, EDNRB, F2R, F2RL1, FFAR1, FFAR2, FFAR4, FPR1, FPR2, FPRL2, GALR2, GCGR, GHSR, GNRHR, GPBAR1, GPM3, GPR35, GPR6, GPR84, GRM1, GRM2, GRM3, GRM4, GRM5, GRM7, HCAR2, HCRTR1, HCRTR2, HRH1, HRH2, HRH3, HRH4, HTR1A, HTR1B, HTR1D, HTR2A, HTR2B, HTR2C, HTR4, HTR5A, HTR6, HTR7, LTB4R, MC1R, MC3R, MC4R, MC5R, MCHR1, MTNR1A, NMUR1, NMUR2, NPBWR1, NPY1R, NPY2R, NTSR1, NTSR2, OPRD1, OPRK1, OPRL1, OPRM1, OXTR, P2RY1, P2RY12, P2RY14, P2RY2, P2RY4, P2RY6, PTAFR, PTGDR, PTGDR2, PTGER1, PTGER2, PTGER3, PTGER4, PTGFR, PTGIR, RXFP1, RXFP2, S1PR1, S1PR3, S1PR4, S1PR5, TAAR1, TACR1, TACR2, TACR3, TBXA2R, TRHR, TSHR |
Ion channels
Ion channel binders constitute important drug groups across a number of diverse therapeutic areas. At the same time, unintended modulation of ion channel activity is a frequent cause of safety issues. Our in silico channel binding assay allows efficient profiling of drug candidates, as well as fast screening for undesired off target activity, including the human Ether-à-go-go-Related Gene (hERG) ion channel, a primary focus for safety testing. The panel includes 137 ion channels (with 66 human targets).
| Assay type | Protein genes |
| Ion channels | CACNA1B, CACNA1C, CACNA1D, CACNA1F, CACNA1G, CACNA1H, CACNA1S, CFTR, CHRNA10, CHRNA2, CHRNA3, CHRNA4, CHRNA5, CHRNA6, CHRNA7, CHRNA9, CHRNB2, CHRNB3, CHRNB4, GABRA1, GABRA2, GABRA3, GABRA4, GABRA5, GABRA6, GABRB1, GABRB2, GABRB3, GABRD, GABRE, GABRG1, GABRG2, GABRG3, GABRP, GABRQ, GRIA1, GRIA2, GRIA3, GRIA4, GRIK1, GRIN1, GRIN2A, GRIN2B, GRIN2C, GRIN2D, GRIN3A, GRIN3B, HTR3A, KCNA3, KCNA5, KCNH2, KCNJ11, KCNJ8, KCNQ2, MCOLN3, NOX1, P2RX3, P2RX4, P2RX7, SCN1A, SCN2A, SCN3A, SCN4A, SCN5A, SCN9A, TRPA1, TRPM2, TRPM8, TRPV1, TRPV2, TRPV4, VPR |
Cytochrome P450
Cytochrome P450 (CYP450) are ubiquitous enzymes responsible for the biotransformation of a wide range of compounds. Accordingly, screening of drug candidates’ interaction with CYP450 enzymes is vital for the assessment of their metabolism, bioavailability, and potential toxicity. Our virtual CYP450 assay covers 12 enzymes, including 11 human proteins.
| Assay type | Protein genes |
| Cytochrome P450 | CYP17A1, CYP19A1, CYP1A1, CYP1A2, CYP1B1, CYP2A6, CYP2C19, CYP2C8, CYP2C9, CYP2D6, CYP3A4, CYP51 |
Nuclear receptors
Nuclear receptors regulate gene expression in response to specific molecular ligands. They are important drug targets in a number of diseases, including hormone-related cancers, metabolic disorders, and autoimmune diseases. The nuclear receptor panel includes 34 unique proteins, with 31 human targets.
| Assay type | Protein genes |
| Nuclear receptors | AR, ESR1, ESR2, ESRRA, ESRRG, HNF4A, HPRT1, NR1I2, NR3C1, NR3C2, PGR, PPARA, PPARD, PPARG, PPIA, RARA, RARB, RARG, RORB, RORC, RXRA, RXRB, RXRG, THRA, THRB, VDR |
Enzymes
Enzymes are among the most important and widely targeted proteins in drug discovery and development due to their central roles in various biochemical pathways and cellular processes. Our panel of virtual binding assays offers possibility for high throughput profiling of enzyme-targetting drug candidates across 597 diverse enzymes, including the most relevant enzyme classes.
| Assay type | Protein genes |
| Oxidoreductases | AKR1B1, AKR1C3, ALOX5, AOC3, CYP17A1, CYP19A1, CYP1A1, CYP1A2, CYP1B1, CYP2A6, CYP2C19, CYP2C8, CYP2C9, CYP2D6, CYP3A4, CYP51, DAO, DHFR, DHOD, DHODH, EGLN1, FABI, FOLA, HMGCR, HPGD, HSD11B1, HSD11B2, HSD17B1, IDO1, IMPDH2, INHA, KDM1A, KDM4A, KDM4C, KDM4E, KDM5A, LDHA, LDHB, MAOA, MAOB, NOS1, NOS2, NOS3, NOX1, NQO1, NQO2, PHGDH, PPO2, PTGS1, PTGS2, SRD5A2, TDO2, TXNRD1, TYR |
| Transferases (in addition to protein kinases) | CARM1, DNMT1, DOT1L, EHMT1, EHMT2, EP300, EZH2, FASN, FDFT1, FDPS, FNTA, FNTB, GGPS1, GSTP1, HPRT1, ICMT, KAT2B, KMT2A, MAT2A, MDM2, NAMPT, NCOA1, NMT1, NNMT, NSD2, PARP1, PARP2, PGGT1B, PNMT, PNP, PRMT1, PRMT5, PRMT6, PYGL, QPCT, SIRT1, SIRT2, SIRT3, SIRT5, SMYD2, SOAT1, TGM2, THYA, TNKS, TNKS2, TYMS, UBE2N, XIAP |
| Hydrolases | ABHD6, ACE, ACHE, ACP1, ADA, ADAM17, ADAMTS5, AHCY, ALPG, ALPI, AMPC, ANPEP, APEX1, APOBEC3A, APOBEC3F, APOBEC3G, ASAH1, ATAD2, BACE1, BACE2, BCHE, BLA, BLA(TEM-2), BLM, C1S, CAPN1, CAPN2, CASP1, CASP3, CASP7, CASP8, CD38, CDC25B, CES2, CMA1, COLA, CPB2, CTSB, CTSD, CTSG, CTSK, CTSL, CTSS, DCTPP1, DEF, DNAB, DPEP1, DPP4, DPP7, DPP8, DPP9, DUSP3, DUSP6, ECE1, ELANE, ENPP2, ERAP1, ERAP2, F10, F11, F12, F2, F7, F9, FAAH, FAP, FBP1, FEN1, GAA, GALC, GBA1, GBA2, GLS, HDAC1, HDAC10, HDAC11, HDAC2, HDAC3, HDAC4, HDAC5, HDAC6, HDAC7, HDAC8, HDAC9, HSP90AA1, KLK1, KLK5, KLK7, KLKB1, KRAS, LEF, LNPEP, LPXC, LTA4H, M18AAP, MALT1, MAP, METAP1, METAP2, MGLL, MME, MMP1, MMP12, MMP13, MMP14, MMP2, MMP3, MMP7, MMP8, MMP9, NA, NAAA, NLRP3, NOTUM, NPR, NT5E, PA, PDE10A, PDE11A, PDE1A, PDE1B, PDE1C, PDE2A, PDE3A, PDE3B, PDE4A, PDE4B, PDE4C, PDE4D, PDE5A, PDE7A, PDE9A, PLA2G1B, PLA2G2A, PLAT, PLAU, PLG, PMII, PNLIP, PPP1CA, PREP, PRSS1, PRSS2, PRSS3, PSEN1, PSEN2, PSMB5, PSMB8, PSMB9, PTPN1, PTPN11, PTPN2, PTPN22, PTPN5, PTPN7, PTPRC, RECQL, REN, SENP6, SENP7, SENP8, SMPD1, STS, STX1, TDP1, TDP2, TMPRSS15, TMPRSS6, TUBA1A, TUBA1B, TUBA1C, TUBA3C, TUBA3E, TUBA4A, USP1, USP19, USP2, USP7, VCP |
| Lyases | CA1, CA12, CA2, CA4, CA5A, CA5B, CA6, CA7, CA9, GLO1 |
| Isomerases | FKBP1A, FKBP5, GYRA, GYRB, MIF, MPI, P4HB, PIN1, PPIA, PTGES, TBXAS1, TOP1, TOP2A, TOP2B |
Transporters
Transporters play a multifaceted role in drug design by influencing drug absorption, distribution, metabolism, and excretion. Understanding transporter-mediated processes and interactions is crucial for optimising drug efficacy, minimising side effects, and ensuring the safe and effective use of medications. Our in siliico transporter screening assay includes 14 targets.
| Assay type | Protein genes |
| Transporters | ABCB1, ABCG2, SLC1A2, SLC29A1, SLC2A1, SLC5A1, SLC5A2, SLC6A1, SLC6A2, SLC6A3, SLC6A4, SLC6A5, SLC6A9, SLCO1B1 |
Classic Computer Aided Drug Design
Molecular modeling
Molecular modeling is a foundation of computer aided drug design (CADD). Initially naive and oversimplified approaches introduced over half a century ago, gradually matured into sophisticated, fully-estabilished tools. In addition to scientific advances, these tools nowadays benefit from the availability of extensive datasets, substantial computational resources, and proliferation of machine learning techniques. Consequently, CADD is now being widely used to accelerate and optimise drug design process across all its stages.
We specialize in the following aspects of molecular modelling:
- biophysical descriptors: 2D, 3D, and physicochemical descriptors, molecular fingerprints
- QM calculations for conformational analysis and force field parameterization of drug-like molecules,
- target protein modelling: homology modelling and refinement of protein structures, binding site annotation, hydration analysis, simulations of protein dynamics & conformational variability,
- lead compound optimisation: simulation-based free energy methods for absolute and relative affinity assessment.
Virtual screening
Virtual High-Throughput Screening (vHTS) aims to predict the biological activity of a large number of compounds against specific targets or pathways based on computer simulation rather than in a physical laboratory setting. Two typical scenarios: structure- or ligand-based vHTS, assume either the knowledge of the target receptor 3D structure or the existence of possibly wide library of small molecules with already measured activity against the desired end-point.
- structure-based vHTS: given 3D model of the receptor (eg. crystallographic of cryo-electron microscope structure) specialised computer software is employed to predict plausible binding modes of the considered compounds as well as their ranking according to calculated binding free energy. We offer an extensive expertise in such docking & scoring campaigns, including:
- assessment and refinement of the receptor model,
- flittering and standardisation of small molecules libraries,
- selection of suitable docking protocol,
- critical evaluation and interpretation of the results.
- ligand-based vHTS: based on the group of molecules whose biological activity in a certain aspect has been quantified, a mathematical relationship is established between their structural and physicochemical properties and the magnitude of the effect under study. The model is subsequently used to estimate the activity of a large pool of screened compounds. Our expertise in such Quantitative Structure-Activity Relationship (QSAR) modelling includes:
- selection and calculation of relevant molecular descriptors,
- formulation of QSAR models using state of the art machine learning approaches,
- thorough evaluation of the results.
ADMET
Evaluation of ADMET (Absorption, Distribution, Metabolism, Excretion, and Toxicity) properties for drug candidates is essential at early stages of development to limit the risk of potential safety and efficacy issues. Calculation of ADMET profile based on molecular structures is a fast and cost efficient method allowing to filter out high-risk compounds from the dataset. We offer the following approaches to the evaluation of ADMET properties:
- descriptor-based toxicity assessment, which relies on machine learning models trained to predict a range of toxicity endpoints based on molecular fingerprints,
- target-based toxicity assessment, which uses molecular docking and structural modeling to assess how a molecule interacts with biological targets and infer potential toxicity mechanisms,
- Physiologically-Based Pharmacokinetic (PBPK) modelling, which takes into account the physiological processes and organ-specific characteristics to construct a mathematical model for predicting drug concentration-time profiles within different body compartments.