Neuraldecipher – reverse-engineering extended-connectivity fingerprints (ECFPs) to their molecular structures

Tuan Le; Robin Winter; Frank Noé; Djork-Arné Clevert

doi:10.1039/d0sc03115a

Neuraldecipher - Reverse-Engineering ECFP Fingerprints to Their Molecular Structures

10.26434/chemrxiv.12286727 ◽

2020 ◽

Author(s):

Tuan Le ◽

Robin Winter ◽

Frank Noé ◽

Djork-Arné Clevert

Keyword(s):

Molecular Structure ◽

Reverse Engineering ◽

Quantitative Structure Activity Relationship ◽

Molecular Structures ◽

Pharmaceutical Companies ◽

Molecular Fingerprints ◽

Structure Activity ◽

Great Relevance ◽

Validation Set ◽

Private Associations

Protecting molecular structures from disclosure against external parties is of great relevance for industrial and private associations, such as pharmaceutical companies. Within the framework of external collaborations, it is common to exchange datasets by encoding the molecular structures into descriptors. Molecular fingerprints such as the extended-connectivity fingerprints are frequently used for such an exchange, because they typically perform well on quantitative structure-activity relationship tasks. ECFPs are often considered to be non-invertible due to the way they are computed.In this paper, we present a reverse-engineering method to deduce the molecular structure given revealed ECFPs. Our method includes the Neuraldecipher, a neural network model that predicts a compact vector representation of compounds, given ECFPs. We then utilize another pre-trained model to retrieve the molecular structure as SMILES representation. We demonstrate that our method is able to reconstruct molecular structures to some extent, and improves, when ECFPs with larger fingerprint sizes are revealed. For example, given ECFP count vectors of length 4096, we are able to correctly deduce around 60% of molecular structures on a validation set (112K unique samples) with our method.

Download Full-text

Neuraldecipher - Reverse-Engineering ECFP Fingerprints to Their Molecular Structures

10.26434/chemrxiv.12286727.v1 ◽

2020 ◽

Author(s):

Tuan Le ◽

Robin Winter ◽

Frank Noé ◽

Djork-Arné Clevert

Keyword(s):

Molecular Structure ◽

Reverse Engineering ◽

Quantitative Structure Activity Relationship ◽

Molecular Structures ◽

Pharmaceutical Companies ◽

Molecular Fingerprints ◽

Structure Activity ◽

Great Relevance ◽

Validation Set ◽

Private Associations

Protecting molecular structures from disclosure against external parties is of great relevance for industrial and private associations, such as pharmaceutical companies. Within the framework of external collaborations, it is common to exchange datasets by encoding the molecular structures into descriptors. Molecular fingerprints such as the extended-connectivity fingerprints are frequently used for such an exchange, because they typically perform well on quantitative structure-activity relationship tasks. ECFPs are often considered to be non-invertible due to the way they are computed.In this paper, we present a reverse-engineering method to deduce the molecular structure given revealed ECFPs. Our method includes the Neuraldecipher, a neural network model that predicts a compact vector representation of compounds, given ECFPs. We then utilize another pre-trained model to retrieve the molecular structure as SMILES representation. We demonstrate that our method is able to reconstruct molecular structures to some extent, and improves, when ECFPs with larger fingerprint sizes are revealed. For example, given ECFP count vectors of length 4096, we are able to correctly deduce around 60% of molecular structures on a validation set (112K unique samples) with our method.

Download Full-text

Neuraldecipher - Reverse-Engineering ECFP Fingerprints to Their Molecular Structures

10.26434/chemrxiv.12286727.v2 ◽

2020 ◽

Author(s):

Tuan Le ◽

Robin Winter ◽

Frank Noé ◽

Djork-Arné Clevert

Keyword(s):

Molecular Structure ◽

Reverse Engineering ◽

Quantitative Structure Activity Relationship ◽

Molecular Structures ◽

Pharmaceutical Companies ◽

Molecular Fingerprints ◽

Structure Activity ◽

Great Relevance ◽

Validation Set ◽

Private Associations

Protecting molecular structures from disclosure against external parties is of great relevance for industrial and private associations, such as pharmaceutical companies. Within the framework of external collaborations, it is common to exchange datasets by encoding the molecular structures into descriptors. Molecular fingerprints such as the extended-connectivity fingerprints are frequently used for such an exchange, because they typically perform well on quantitative structure-activity relationship tasks. ECFPs are often considered to be non-invertible due to the way they are computed.In this paper, we present a reverse-engineering method to deduce the molecular structure given revealed ECFPs. Our method includes the Neuraldecipher, a neural network model that predicts a compact vector representation of compounds, given ECFPs. We then utilize another pre-trained model to retrieve the molecular structure as SMILES representation. We demonstrate that our method is able to reconstruct molecular structures to some extent, and improves, when ECFPs with larger fingerprint sizes are revealed. For example, given ECFP count vectors of length 4096, we are able to correctly deduce around 60% of molecular structures on a validation set (112K unique samples) with our method.

Download Full-text

Electron Microscopy in Molecular Biology

Proceedings, annual meeting, Electron Microscopy Society of America ◽

10.1017/s0424820100060027 ◽

1967 ◽

Vol 25 ◽

pp. 2-3

Author(s):

Cecil E. Hall

Keyword(s):

Electron Microscopy ◽

Molecular Biology ◽

Noise Level ◽

Molecular Structures ◽

Material Structure ◽

The Arts ◽

Shadow Casting ◽

Electron Image ◽

High Degree ◽

Very High

The visualization of organic macromolecules such as proteins, nucleic acids, viruses and virus components has reached its high degree of effectiveness owing to refinements and reliability of instruments and to the invention of methods for enhancing the structure of these materials within the electron image. The latter techniques have been most important because what can be seen depends upon the molecular and atomic character of the object as modified which is rarely evident in the pristine material. Structure may thus be displayed by the arts of positive and negative staining, shadow casting, replication and other techniques. Enhancement of contrast, which delineates bounds of isolated macromolecules has been effected progressively over the years as illustrated in Figs. 1, 2, 3 and 4 by these methods. We now look to the future wondering what other visions are waiting to be seen. The instrument designers will need to exact from the arts of fabrication the performance that theory has prescribed as well as methods for phase and interference contrast with explorations of the potentialities of very high and very low voltages. Chemistry must play an increasingly important part in future progress by providing specific stain molecules of high visibility, substrates of vanishing “noise” level and means for preservation of molecular structures that usually exist in a solvated condition.

Download Full-text

Scanning tunneling microscopy (STM) of DNA and RNA

Proceedings, annual meeting, Electron Microscopy Society of America ◽

10.1017/s0424820100152148 ◽

1989 ◽

Vol 47 ◽

pp. 34-35

Author(s):

Patricia G. Arscott ◽

Gil Lee ◽

Victor A. Bloomfield ◽

D. Fennell Evans

Keyword(s):

Molecular Structures ◽

Inorganic Materials ◽

Resolving Power ◽

Scanning Tunneling ◽

Tunneling Microscopy ◽

Form And Function ◽

Dna And Rna ◽

Calf Thymus ◽

And Function ◽

The Relationship

STM is one of the most promising techniques available for visualizing the fine details of biomolecular structure. It has been used to map the surface topography of inorganic materials in atomic dimensions, and thus has the resolving power not only to determine the conformation of small molecules but to distinguish site-specific features within a molecule. That level of detail is of critical importance in understanding the relationship between form and function in biological systems. The size, shape, and accessibility of molecular structures can be determined much more accurately by STM than by electron microscopy since no staining, shadowing or labeling with heavy metals is required, and there is no exposure to damaging radiation by electrons. Crystallography and most other physical techniques do not give information about individual molecules.We have obtained striking images of DNA and RNA, using calf thymus DNA and two synthetic polynucleotides, poly(dG-me5dC)·poly(dG-me5dC) and poly(rA)·poly(rU).

Download Full-text

Molecular architecture and functions of the neuronal cytoskeleton

Proceedings, annual meeting, Electron Microscopy Society of America ◽

10.1017/s0424820100157541 ◽

1990 ◽

Vol 48 (3) ◽

pp. 2-3

Author(s):

Nobutaka Hirokawa

Keyword(s):

Major Elements ◽

Molecular Structures ◽

Organelle Transport ◽

Molecular Architecture ◽

Neuronal Morphogenesis ◽

Microtubule Associated Proteins ◽

Associated Proteins ◽

And Function ◽

Small Clusters

In this symposium I will present our studies about the molecular architecture and function of the cytomatrix of the nerve cells. The nerve cell is a highly polarized cell composed of highly branched dendrites, cell body, and a single long axon along the direction of the impulse propagation. Each part of the neuron takes characteristic shapes for which the cytoskeleton provides the framework. The neuronal cytoskeletons play important roles on neuronal morphogenesis, organelle transport and the synaptic transmission. In the axon neurofilaments (NF) form dense arrays, while microtubules (MT) are arranged as small clusters among the NFs. On the other hand, MTs are distributed uniformly, whereas NFs tend to run solitarily or form small fascicles in the dendrites Quick freeze deep etch electron microscopy revealed various kinds of strands among MTs, NFs and membranous organelles (MO). These structures form major elements of the cytomatrix in the neuron. To investigate molecular nature and function of these filaments first we studied molecular structures of microtubule associated proteins (MAP1A, MAP1B, MAP2, MAP2C and tau), and microtubules reconstituted from MAPs and tubulin in vitro. These MAPs were all fibrous molecules with different length and formed arm like projections from the microtubule surface.

Download Full-text

The Eleventh Circuit Holds that Agreements in which Pharmaceutical Companies Pay Generic Companies Not to Compete May be Valid

The Journal of Law Medicine & Ethics ◽

10.1017/s107311050000749x ◽

2004 ◽

Vol 32 (1) ◽

pp. 181-184

Author(s):

Amy Garrigues

Keyword(s):

Pharmaceutical Companies ◽

Court Decisions ◽

Patent Holder ◽

Trial Court ◽

Court Of Appeals ◽

The Us ◽

Antitrust Laws ◽

Per Se ◽

Appeals Court ◽

Two Sides

On September 15, 2003, the US. Court of Appeals for the Eleventh Circuit held that agreements between pharmaceutical and generic companies not to compete are not per se unlawful if these agreements do not expand the existing exclusionary right of a patent. The Valley DrugCo.v.Geneva Pharmaceuticals decision emphasizes that the nature of a patent gives the patent holder exclusive rights, and if an agreement merely confirms that exclusivity, then it is not per se unlawful. With this holding, the appeals court reversed the decision of the trial court, which held that agreements under which competitors are paid to stay out of the market are per se violations of the antitrust laws. An examination of the Valley Drugtrial and appeals court decisions sheds light on the two sides of an emerging legal debate concerning the validity of pay-not-to-compete agreements, and more broadly, on the appropriate balance between the seemingly competing interests of patent and antitrust laws.

Download Full-text